From d26c7ac48ded2f07e9e934cc4788239987bb2754 Mon Sep 17 00:00:00 2001
From: Keegan George <kgeorge13@gmail.com>
Date: Thu, 17 Apr 2025 15:09:48 -0700
Subject: [PATCH] FEATURE: Add spending metrics to AI usage (#1268)

This update adds metrics for estimated spending in AI usage. To make use of it, admins must add cost details to the LLM config page (input, output, and cached input costs per 1M tokens). After doing so, the metrics will appear in the AI usage dashboard as the AI plugin is used.
---
 ...min-plugins-show-discourse-ai-llms-edit.js |  11 +-
 .../discourse_ai/admin/ai_llms_controller.rb  |   3 +
 app/models/llm_model.rb                       |  12 +-
 app/serializers/ai_usage_serializer.rb        |  10 +
 app/serializers/llm_model_serializer.rb       |   3 +
 .../discourse/admin/models/ai-llm.js          |   5 +-
 .../components/ai-llm-editor-form.gjs         | 512 +++++++++---------
 .../discourse/components/ai-usage.gjs         |  61 ++-
 config/locales/client.en.yml                  |  10 +
 ...416215039_add_cost_metrics_to_llm_model.rb |   9 +
 lib/completions/llm.rb                        |  72 ++-
 lib/completions/report.rb                     |  74 ++-
 spec/fabricators/llm_model_fabricator.rb      |   3 +
 .../admin/ai_usage_controller_spec.rb         |  35 +-
 14 files changed, 559 insertions(+), 261 deletions(-)
 create mode 100644 db/migrate/20250416215039_add_cost_metrics_to_llm_model.rb

diff --git a/admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-llms-edit.js b/admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-llms-edit.js
index 8230211c..f7b5de4c 100644
--- a/admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-llms-edit.js
+++ b/admin/assets/javascripts/discourse/routes/admin-plugins-show-discourse-ai-llms-edit.js
@@ -2,8 +2,17 @@ import DiscourseRoute from "discourse/routes/discourse";
 
 export default class AdminPluginsShowDiscourseAiLlmsEdit extends DiscourseRoute {
   async model(params) {
-    const allLlms = this.modelFor("adminPlugins.show.discourse-ai-llms");
     const id = parseInt(params.id, 10);
+
+    if (id < 0) {
+      // You shouldn't be able to access the edit page
+      // if the model is seeded
+      return this.router.transitionTo(
+        "adminPlugins.show.discourse-ai-llms.index"
+      );
+    }
+
+    const allLlms = this.modelFor("adminPlugins.show.discourse-ai-llms");
     const record = allLlms.findBy("id", id);
     record.provider_params = record.provider_params || {};
     return record;
diff --git a/app/controllers/discourse_ai/admin/ai_llms_controller.rb b/app/controllers/discourse_ai/admin/ai_llms_controller.rb
index 68ddcf94..84109220 100644
--- a/app/controllers/discourse_ai/admin/ai_llms_controller.rb
+++ b/app/controllers/discourse_ai/admin/ai_llms_controller.rb
@@ -161,6 +161,9 @@ module DiscourseAi
             :api_key,
             :enabled_chat_bot,
             :vision_enabled,
+            :input_cost,
+            :cached_input_cost,
+            :output_cost,
           )
 
         provider = updating ? updating.provider : permitted[:provider]
diff --git a/app/models/llm_model.rb b/app/models/llm_model.rb
index 111ac70a..05c7be4b 100644
--- a/app/models/llm_model.rb
+++ b/app/models/llm_model.rb
@@ -13,7 +13,14 @@ class LlmModel < ActiveRecord::Base
   validates :url, presence: true, unless: -> { provider == BEDROCK_PROVIDER_NAME }
   validates_presence_of :name, :api_key
   validates :max_prompt_tokens, numericality: { greater_than: 0 }
-  validates :max_output_tokens, numericality: { greater_than_or_equal_to: 0 }, allow_nil: true
+  validates :input_cost,
+            :cached_input_cost,
+            :output_cost,
+            :max_output_tokens,
+            numericality: {
+              greater_than_or_equal_to: 0,
+            },
+            allow_nil: true
   validate :required_provider_params
   scope :in_use,
         -> do
@@ -184,5 +191,8 @@ end
 #  enabled_chat_bot  :boolean          default(FALSE), not null
 #  provider_params   :jsonb
 #  vision_enabled    :boolean          default(FALSE), not null
+#  input_cost        :float
+#  cached_input_cost :float
+#  output_cost       :float
 #  max_output_tokens :integer
 #
diff --git a/app/serializers/ai_usage_serializer.rb b/app/serializers/ai_usage_serializer.rb
index 38969b05..0a990547 100644
--- a/app/serializers/ai_usage_serializer.rb
+++ b/app/serializers/ai_usage_serializer.rb
@@ -22,6 +22,9 @@ class AiUsageSerializer < ApplicationSerializer
         total_cached_tokens
         total_request_tokens
         total_response_tokens
+        input_spending
+        output_spending
+        cached_input_spending
       ],
     )
   end
@@ -35,6 +38,9 @@ class AiUsageSerializer < ApplicationSerializer
         total_cached_tokens
         total_request_tokens
         total_response_tokens
+        input_spending
+        output_spending
+        cached_input_spending
       ],
     )
   end
@@ -49,6 +55,9 @@ class AiUsageSerializer < ApplicationSerializer
         total_cached_tokens: user.total_cached_tokens,
         total_request_tokens: user.total_request_tokens,
         total_response_tokens: user.total_response_tokens,
+        input_spending: user.input_spending,
+        output_spending: user.output_spending,
+        cached_input_spending: user.cached_input_spending,
       }
     end
   end
@@ -60,6 +69,7 @@ class AiUsageSerializer < ApplicationSerializer
       total_request_tokens: object.total_request_tokens,
       total_response_tokens: object.total_response_tokens,
       total_requests: object.total_requests,
+      total_spending: object.total_spending,
       date_range: {
         start: object.start_date,
         end: object.end_date,
diff --git a/app/serializers/llm_model_serializer.rb b/app/serializers/llm_model_serializer.rb
index 33e62d8b..736757d4 100644
--- a/app/serializers/llm_model_serializer.rb
+++ b/app/serializers/llm_model_serializer.rb
@@ -18,6 +18,9 @@ class LlmModelSerializer < ApplicationSerializer
              :enabled_chat_bot,
              :provider_params,
              :vision_enabled,
+             :input_cost,
+             :output_cost,
+             :cached_input_cost,
              :used_by
 
   has_one :user, serializer: BasicUserSerializer, embed: :object
diff --git a/assets/javascripts/discourse/admin/models/ai-llm.js b/assets/javascripts/discourse/admin/models/ai-llm.js
index c434ba25..cafbf3a0 100644
--- a/assets/javascripts/discourse/admin/models/ai-llm.js
+++ b/assets/javascripts/discourse/admin/models/ai-llm.js
@@ -15,7 +15,10 @@ export default class AiLlm extends RestModel {
       "api_key",
       "enabled_chat_bot",
       "provider_params",
-      "vision_enabled"
+      "vision_enabled",
+      "input_cost",
+      "cached_input_cost",
+      "output_cost"
     );
   }
 
diff --git a/assets/javascripts/discourse/components/ai-llm-editor-form.gjs b/assets/javascripts/discourse/components/ai-llm-editor-form.gjs
index 50ab2ff7..38cf1b50 100644
--- a/assets/javascripts/discourse/components/ai-llm-editor-form.gjs
+++ b/assets/javascripts/discourse/components/ai-llm-editor-form.gjs
@@ -47,6 +47,9 @@ export default class AiLlmEditorForm extends Component {
         name: modelInfo.name,
         provider: info.provider,
         provider_params: this.computeProviderParams(info.provider),
+        input_cost: modelInfo.input_cost,
+        output_cost: modelInfo.output_cost,
+        cached_input_cost: modelInfo.cached_input_cost,
       };
     }
 
@@ -63,6 +66,9 @@ export default class AiLlmEditorForm extends Component {
       provider: model.provider,
       enabled_chat_bot: model.enabled_chat_bot,
       vision_enabled: model.vision_enabled,
+      input_cost: model.input_cost,
+      output_cost: model.output_cost,
+      cached_input_cost: model.cached_input_cost,
       provider_params: this.computeProviderParams(
         model.provider,
         model.provider_params
@@ -118,10 +124,6 @@ export default class AiLlmEditorForm extends Component {
     return localized.join(", ");
   }
 
-  get seeded() {
-    return this.args.model.id < 0;
-  }
-
   get inUseWarning() {
     return i18n("discourse_ai.llms.in_use_warning", {
       settings: this.modulesUsingModel,
@@ -271,15 +273,9 @@ export default class AiLlmEditorForm extends Component {
     <Form
       @onSubmit={{this.save}}
       @data={{this.formData}}
-      class="ai-llm-editor {{if this.seeded 'seeded'}}"
+      class="ai-llm-editor"
       as |form data|
     >
-      {{#if this.seeded}}
-        <form.Alert @icon="circle-info">
-          {{i18n "discourse_ai.llms.seeded_warning"}}
-        </form.Alert>
-      {{/if}}
-
       {{#if this.modulesUsingModel}}
         <form.Alert @icon="circle-info">
           {{this.inUseWarning}}
@@ -290,7 +286,6 @@ export default class AiLlmEditorForm extends Component {
         @name="display_name"
         @title={{i18n "discourse_ai.llms.display_name"}}
         @validation="required|length:1,100"
-        @disabled={{this.seeded}}
         @format="large"
         @tooltip={{i18n "discourse_ai.llms.hints.display_name"}}
         as |field|
@@ -303,7 +298,6 @@ export default class AiLlmEditorForm extends Component {
         @title={{i18n "discourse_ai.llms.name"}}
         @tooltip={{i18n "discourse_ai.llms.hints.name"}}
         @validation="required"
-        @disabled={{this.seeded}}
         @format="large"
         as |field|
       >
@@ -313,7 +307,6 @@ export default class AiLlmEditorForm extends Component {
       <form.Field
         @name="provider"
         @title={{i18n "discourse_ai.llms.provider"}}
-        @disabled={{this.seeded}}
         @format="large"
         @validation="required"
         @onSet={{this.setProvider}}
@@ -328,262 +321,289 @@ export default class AiLlmEditorForm extends Component {
         </field.Select>
       </form.Field>
 
-      {{#unless this.seeded}}
-        {{#if (this.canEditURL data.provider)}}
-          <form.Field
-            @name="url"
-            @title={{i18n "discourse_ai.llms.url"}}
-            @validation="required"
-            @format="large"
-            as |field|
-          >
-            <field.Input />
-          </form.Field>
-        {{/if}}
-
+      {{#if (this.canEditURL data.provider)}}
         <form.Field
-          @name="api_key"
-          @title={{i18n "discourse_ai.llms.api_key"}}
+          @name="url"
+          @title={{i18n "discourse_ai.llms.url"}}
           @validation="required"
           @format="large"
           as |field|
         >
-          <field.Password />
+          <field.Input />
         </form.Field>
+      {{/if}}
 
-        <form.Object @name="provider_params" as |object providerParamsData|>
-          {{#each (this.providerParamsKeys providerParamsData) as |name|}}
-            {{#let
-              (get (this.metaProviderParams data.provider) name)
-              as |params|
-            }}
-              <object.Field
-                @name={{name}}
-                @title={{i18n
-                  (concat "discourse_ai.llms.provider_fields." name)
-                }}
-                @format="large"
-                as |field|
-              >
-                {{#if (eq params.type "enum")}}
-                  <field.Select @includeNone={{false}} as |select|>
-                    {{#each params.values as |option|}}
-                      <select.Option
-                        @value={{option.id}}
-                      >{{option.name}}</select.Option>
-                    {{/each}}
-                  </field.Select>
-                {{else if (eq params.type "checkbox")}}
-                  <field.Checkbox />
-                {{else}}
-                  <field.Input @type={{params.type}} />
-                {{/if}}
-              </object.Field>
-            {{/let}}
-          {{/each}}
-        </form.Object>
+      <form.Field
+        @name="api_key"
+        @title={{i18n "discourse_ai.llms.api_key"}}
+        @validation="required"
+        @format="large"
+        as |field|
+      >
+        <field.Password />
+      </form.Field>
 
-        <form.Field
-          @name="tokenizer"
-          @title={{i18n "discourse_ai.llms.tokenizer"}}
-          @disabled={{this.seeded}}
-          @format="large"
-          @validation="required"
-          as |field|
-        >
-          <field.Select as |select|>
-            {{#each this.tokenizers as |tokenizer|}}
-              <select.Option
-                @value={{tokenizer.id}}
-              >{{tokenizer.name}}</select.Option>
-            {{/each}}
-          </field.Select>
-        </form.Field>
-
-        <form.Field
-          @name="max_prompt_tokens"
-          @title={{i18n "discourse_ai.llms.max_prompt_tokens"}}
-          @tooltip={{i18n "discourse_ai.llms.hints.max_prompt_tokens"}}
-          @validation="required"
-          @format="large"
-          as |field|
-        >
-          <field.Input @type="number" step="any" min="0" lang="en" />
-        </form.Field>
-
-        <form.Field
-          @name="max_output_tokens"
-          @title={{i18n "discourse_ai.llms.max_output_tokens"}}
-          @tooltip={{i18n "discourse_ai.llms.hints.max_output_tokens"}}
-          @format="large"
-          as |field|
-        >
-          <field.Input @type="number" step="any" min="0" lang="en" />
-        </form.Field>
-
-        <form.Field
-          @name="vision_enabled"
-          @title={{i18n "discourse_ai.llms.vision_enabled"}}
-          @tooltip={{i18n "discourse_ai.llms.hints.vision_enabled"}}
-          @format="large"
-          as |field|
-        >
-          <field.Checkbox />
-        </form.Field>
-
-        <form.Field
-          @name="enabled_chat_bot"
-          @title={{i18n "discourse_ai.llms.enabled_chat_bot"}}
-          @tooltip={{i18n "discourse_ai.llms.hints.enabled_chat_bot"}}
-          @format="large"
-          as |field|
-        >
-          <field.Checkbox />
-        </form.Field>
-
-        {{#if @model.user}}
-          <form.Container @title={{i18n "discourse_ai.llms.ai_bot_user"}}>
-            <a
-              class="avatar"
-              href={{@model.user.path}}
-              data-user-card={{@model.user.username}}
+      <form.Object @name="provider_params" as |object providerParamsData|>
+        {{#each (this.providerParamsKeys providerParamsData) as |name|}}
+          {{#let
+            (get (this.metaProviderParams data.provider) name)
+            as |params|
+          }}
+            <object.Field
+              @name={{name}}
+              @title={{i18n (concat "discourse_ai.llms.provider_fields." name)}}
+              @format="large"
+              as |field|
             >
-              {{Avatar @model.user.avatar_template "small"}}
-            </a>
-            <LinkTo @route="adminUser" @model={{this.adminUser}}>
-              {{@model.user.username}}
-            </LinkTo>
-          </form.Container>
-        {{/if}}
+              {{#if (eq params.type "enum")}}
+                <field.Select @includeNone={{false}} as |select|>
+                  {{#each params.values as |option|}}
+                    <select.Option
+                      @value={{option.id}}
+                    >{{option.name}}</select.Option>
+                  {{/each}}
+                </field.Select>
+              {{else if (eq params.type "checkbox")}}
+                <field.Checkbox />
+              {{else}}
+                <field.Input @type={{params.type}} />
+              {{/if}}
+            </object.Field>
+          {{/let}}
+        {{/each}}
+      </form.Object>
 
-        {{#if (gt data.llm_quotas.length 0)}}
-          <form.Container @title={{i18n "discourse_ai.llms.quotas.title"}}>
-            <table class="ai-llm-quotas__table">
-              <thead class="ai-llm-quotas__table-head">
-                <tr class="ai-llm-quotas__header-row">
-                  <th class="ai-llm-quotas__header">{{i18n
-                      "discourse_ai.llms.quotas.group"
-                    }}</th>
-                  <th class="ai-llm-quotas__header">{{i18n
-                      "discourse_ai.llms.quotas.max_tokens"
-                    }}</th>
-                  <th class="ai-llm-quotas__header">{{i18n
-                      "discourse_ai.llms.quotas.max_usages"
-                    }}</th>
-                  <th class="ai-llm-quotas__header">{{i18n
-                      "discourse_ai.llms.quotas.duration"
-                    }}</th>
-                  <th
-                    class="ai-llm-quotas__header ai-llm-quotas__header--actions"
-                  ></th>
-                  <th></th>
-                </tr>
-              </thead>
-              <tbody class="ai-llm-quotas__table-body">
-                <form.Collection
-                  @name="llm_quotas"
-                  @tagName="tr"
-                  class="ai-llm-quotas__row"
-                  as |collection index collectionData|
-                >
-                  <td
-                    class="ai-llm-quotas__cell"
-                  >{{collectionData.group_name}}</td>
-                  <td class="ai-llm-quotas__cell">
-                    <collection.Field
-                      @name="max_tokens"
-                      @title="max_tokens"
-                      @showTitle={{false}}
-                      as |field|
-                    >
-                      <field.Input
-                        @type="number"
-                        class="ai-llm-quotas__input"
-                        min="1"
-                      />
-                    </collection.Field>
-                  </td>
-                  <td class="ai-llm-quotas__cell">
-                    <collection.Field
-                      @name="max_usages"
-                      @title="max_usages"
-                      @showTitle={{false}}
-                      as |field|
-                    >
-                      <field.Input
-                        @type="number"
-                        class="ai-llm-quotas__input"
-                        min="1"
-                      />
-                    </collection.Field>
-                  </td>
-                  <td class="ai-llm-quotas__cell">
-                    <collection.Field
-                      @name="duration_seconds"
-                      @title="duration_seconds"
-                      @showTitle={{false}}
-                      as |field|
-                    >
-                      <field.Custom>
-                        <DurationSelector
-                          @value={{collectionData.duration_seconds}}
-                          @onChange={{field.set}}
-                        />
-                      </field.Custom>
-                    </collection.Field>
-                  </td>
-                  <td>
-                    <form.Button
-                      @icon="trash-can"
-                      @action={{fn collection.remove index}}
-                      class="btn-danger ai-llm-quotas__delete-btn"
+      <form.Field
+        @name="tokenizer"
+        @title={{i18n "discourse_ai.llms.tokenizer"}}
+        @format="large"
+        @validation="required"
+        as |field|
+      >
+        <field.Select as |select|>
+          {{#each this.tokenizers as |tokenizer|}}
+            <select.Option
+              @value={{tokenizer.id}}
+            >{{tokenizer.name}}</select.Option>
+          {{/each}}
+        </field.Select>
+      </form.Field>
+
+      <form.Field
+        @name="max_prompt_tokens"
+        @title={{i18n "discourse_ai.llms.max_prompt_tokens"}}
+        @tooltip={{i18n "discourse_ai.llms.hints.max_prompt_tokens"}}
+        @validation="required"
+        @format="large"
+        as |field|
+      >
+        <field.Input @type="number" step="any" min="0" lang="en" />
+      </form.Field>
+
+      <form.InputGroup as |inputGroup|>
+        <inputGroup.Field
+          @name="input_cost"
+          @title={{i18n "discourse_ai.llms.cost_input"}}
+          @tooltip={{i18n "discourse_ai.llms.hints.cost_input"}}
+          @helpText={{i18n "discourse_ai.llms.hints.cost_measure"}}
+          as |field|
+        >
+          <field.Input @type="number" step="any" min="0" lang="en" />
+        </inputGroup.Field>
+
+        <inputGroup.Field
+          @name="cached_input_cost"
+          @title={{i18n "discourse_ai.llms.cost_cached_input"}}
+          @tooltip={{i18n "discourse_ai.llms.hints.cost_cached_input"}}
+          @helpText={{i18n "discourse_ai.llms.hints.cost_measure"}}
+          as |field|
+        >
+          <field.Input @type="number" step="any" min="0" lang="en" />
+        </inputGroup.Field>
+
+        <inputGroup.Field
+          @name="output_cost"
+          @title={{i18n "discourse_ai.llms.cost_output"}}
+          @tooltip={{i18n "discourse_ai.llms.hints.cost_output"}}
+          @helpText={{i18n "discourse_ai.llms.hints.cost_measure"}}
+          as |field|
+        >
+          <field.Input @type="number" step="any" min="0" lang="en" />
+        </inputGroup.Field>
+      </form.InputGroup>
+
+      <form.Field
+        @name="max_output_tokens"
+        @title={{i18n "discourse_ai.llms.max_output_tokens"}}
+        @tooltip={{i18n "discourse_ai.llms.hints.max_output_tokens"}}
+        @format="large"
+        as |field|
+      >
+        <field.Input @type="number" step="any" min="0" lang="en" />
+      </form.Field>
+
+      <form.Field
+        @name="vision_enabled"
+        @title={{i18n "discourse_ai.llms.vision_enabled"}}
+        @tooltip={{i18n "discourse_ai.llms.hints.vision_enabled"}}
+        @format="large"
+        as |field|
+      >
+        <field.Checkbox />
+      </form.Field>
+
+      <form.Field
+        @name="enabled_chat_bot"
+        @title={{i18n "discourse_ai.llms.enabled_chat_bot"}}
+        @tooltip={{i18n "discourse_ai.llms.hints.enabled_chat_bot"}}
+        @format="large"
+        as |field|
+      >
+        <field.Checkbox />
+      </form.Field>
+
+      {{#if @model.user}}
+        <form.Container @title={{i18n "discourse_ai.llms.ai_bot_user"}}>
+          <a
+            class="avatar"
+            href={{@model.user.path}}
+            data-user-card={{@model.user.username}}
+          >
+            {{Avatar @model.user.avatar_template "small"}}
+          </a>
+          <LinkTo @route="adminUser" @model={{this.adminUser}}>
+            {{@model.user.username}}
+          </LinkTo>
+        </form.Container>
+      {{/if}}
+
+      {{#if (gt data.llm_quotas.length 0)}}
+        <form.Container @title={{i18n "discourse_ai.llms.quotas.title"}}>
+          <table class="ai-llm-quotas__table">
+            <thead class="ai-llm-quotas__table-head">
+              <tr class="ai-llm-quotas__header-row">
+                <th class="ai-llm-quotas__header">{{i18n
+                    "discourse_ai.llms.quotas.group"
+                  }}</th>
+                <th class="ai-llm-quotas__header">{{i18n
+                    "discourse_ai.llms.quotas.max_tokens"
+                  }}</th>
+                <th class="ai-llm-quotas__header">{{i18n
+                    "discourse_ai.llms.quotas.max_usages"
+                  }}</th>
+                <th class="ai-llm-quotas__header">{{i18n
+                    "discourse_ai.llms.quotas.duration"
+                  }}</th>
+                <th
+                  class="ai-llm-quotas__header ai-llm-quotas__header--actions"
+                ></th>
+                <th></th>
+              </tr>
+            </thead>
+            <tbody class="ai-llm-quotas__table-body">
+              <form.Collection
+                @name="llm_quotas"
+                @tagName="tr"
+                class="ai-llm-quotas__row"
+                as |collection index collectionData|
+              >
+                <td
+                  class="ai-llm-quotas__cell"
+                >{{collectionData.group_name}}</td>
+                <td class="ai-llm-quotas__cell">
+                  <collection.Field
+                    @name="max_tokens"
+                    @title="max_tokens"
+                    @showTitle={{false}}
+                    as |field|
+                  >
+                    <field.Input
+                      @type="number"
+                      class="ai-llm-quotas__input"
+                      min="1"
                     />
-                  </td>
-                </form.Collection>
-              </tbody>
-            </table>
-          </form.Container>
+                  </collection.Field>
+                </td>
+                <td class="ai-llm-quotas__cell">
+                  <collection.Field
+                    @name="max_usages"
+                    @title="max_usages"
+                    @showTitle={{false}}
+                    as |field|
+                  >
+                    <field.Input
+                      @type="number"
+                      class="ai-llm-quotas__input"
+                      min="1"
+                    />
+                  </collection.Field>
+                </td>
+                <td class="ai-llm-quotas__cell">
+                  <collection.Field
+                    @name="duration_seconds"
+                    @title="duration_seconds"
+                    @showTitle={{false}}
+                    as |field|
+                  >
+                    <field.Custom>
+                      <DurationSelector
+                        @value={{collectionData.duration_seconds}}
+                        @onChange={{field.set}}
+                      />
+                    </field.Custom>
+                  </collection.Field>
+                </td>
+                <td>
+                  <form.Button
+                    @icon="trash-can"
+                    @action={{fn collection.remove index}}
+                    class="btn-danger ai-llm-quotas__delete-btn"
+                  />
+                </td>
+              </form.Collection>
+            </tbody>
+          </table>
+        </form.Container>
 
+        <form.Button
+          @action={{fn
+            this.openAddQuotaModal
+            (fn form.addItemToCollection "llm_quotas")
+          }}
+          @icon="plus"
+          @label="discourse_ai.llms.quotas.add"
+          class="ai-llm-editor__add-quota-btn"
+        />
+      {{/if}}
+
+      <form.Actions>
+        <form.Button
+          @action={{fn this.test data}}
+          @disabled={{this.testRunning}}
+          @label="discourse_ai.llms.tests.title"
+        />
+
+        <form.Submit />
+
+        {{#if (eq data.llm_quotas.length 0)}}
           <form.Button
             @action={{fn
               this.openAddQuotaModal
               (fn form.addItemToCollection "llm_quotas")
             }}
-            @icon="plus"
             @label="discourse_ai.llms.quotas.add"
             class="ai-llm-editor__add-quota-btn"
           />
         {{/if}}
 
-        <form.Actions>
+        {{#unless @model.isNew}}
           <form.Button
-            @action={{fn this.test data}}
-            @disabled={{this.testRunning}}
-            @label="discourse_ai.llms.tests.title"
+            @action={{this.delete}}
+            @label="discourse_ai.llms.delete"
+            class="btn-danger"
           />
-
-          <form.Submit />
-
-          {{#if (eq data.llm_quotas.length 0)}}
-            <form.Button
-              @action={{fn
-                this.openAddQuotaModal
-                (fn form.addItemToCollection "llm_quotas")
-              }}
-              @label="discourse_ai.llms.quotas.add"
-              class="ai-llm-editor__add-quota-btn"
-            />
-          {{/if}}
-
-          {{#unless @model.isNew}}
-            <form.Button
-              @action={{this.delete}}
-              @label="discourse_ai.llms.delete"
-              class="btn-danger"
-            />
-          {{/unless}}
-        </form.Actions>
-      {{/unless}}
+        {{/unless}}
+      </form.Actions>
 
       {{#if this.displayTestResult}}
         <form.Container @format="full">
diff --git a/assets/javascripts/discourse/components/ai-usage.gjs b/assets/javascripts/discourse/components/ai-usage.gjs
index a972726b..0aaac6b2 100644
--- a/assets/javascripts/discourse/components/ai-usage.gjs
+++ b/assets/javascripts/discourse/components/ai-usage.gjs
@@ -2,6 +2,8 @@ import Component from "@glimmer/component";
 import { tracked } from "@glimmer/tracking";
 import { fn, hash } from "@ember/helper";
 import { action } from "@ember/object";
+import didInsert from "@ember/render-modifiers/modifiers/did-insert";
+import didUpdate from "@ember/render-modifiers/modifiers/did-update";
 import { LinkTo } from "@ember/routing";
 import { service } from "@ember/service";
 import { eq, gt, lt } from "truth-helpers";
@@ -74,6 +76,22 @@ export default class AiUsage extends Component {
     this.onFilterChange();
   }
 
+  @action
+  addCurrencyChar(element) {
+    element.querySelectorAll(".d-stat-tile__label").forEach((label) => {
+      if (
+        label.innerText.trim() === i18n("discourse_ai.usage.total_spending")
+      ) {
+        const valueElement = label
+          .closest(".d-stat-tile")
+          ?.querySelector(".d-stat-tile__value");
+        if (valueElement) {
+          valueElement.innerText = `$${valueElement.innerText}`;
+        }
+      }
+    });
+  }
+
   @bind
   takeUsers(start, end) {
     return this.data.users.slice(start, end);
@@ -153,6 +171,11 @@ export default class AiUsage extends Component {
         value: this.data.summary.total_cached_tokens,
         tooltip: i18n("discourse_ai.usage.stat_tooltips.cached_tokens"),
       },
+      {
+        label: i18n("discourse_ai.usage.total_spending"),
+        value: this.data.summary.total_spending,
+        tooltip: i18n("discourse_ai.usage.stat_tooltips.total_spending"),
+      },
     ];
   }
 
@@ -308,6 +331,11 @@ export default class AiUsage extends Component {
     this.fetchData();
   }
 
+  totalSpending(inputSpending, cachedSpending, outputSpending) {
+    const total = inputSpending + cachedSpending + outputSpending;
+    return `$${total.toFixed(2)}`;
+  }
+
   <template>
     <div class="ai-usage admin-detail">
       <DPageSubheader
@@ -376,9 +404,15 @@ export default class AiUsage extends Component {
             class="ai-usage__summary"
           >
             <:content>
-              <DStatTiles as |tiles|>
+              <DStatTiles
+                {{didInsert this.addCurrencyChar this.metrics}}
+                {{didUpdate this.addCurrencyChar this.metrics}}
+                as |tiles|
+              >
+
                 {{#each this.metrics as |metric|}}
                   <tiles.Tile
+                    class="bar"
                     @label={{metric.label}}
                     @href={{metric.href}}
                     @value={{metric.value}}
@@ -422,6 +456,7 @@ export default class AiUsage extends Component {
                         <th>{{i18n "discourse_ai.usage.feature"}}</th>
                         <th>{{i18n "discourse_ai.usage.usage_count"}}</th>
                         <th>{{i18n "discourse_ai.usage.total_tokens"}}</th>
+                        <th>{{i18n "discourse_ai.usage.total_spending"}}</th>
                       </tr>
                     </thead>
                     <tbody>
@@ -438,6 +473,13 @@ export default class AiUsage extends Component {
                             class="ai-usage__features-cell"
                             title={{feature.total_tokens}}
                           >{{number feature.total_tokens}}</td>
+                          <td>
+                            {{this.totalSpending
+                              feature.input_spending
+                              feature.cached_input_spending
+                              feature.output_spending
+                            }}
+                          </td>
                         </tr>
                       {{/each}}
                     </tbody>
@@ -464,6 +506,8 @@ export default class AiUsage extends Component {
                         <th>{{i18n "discourse_ai.usage.model"}}</th>
                         <th>{{i18n "discourse_ai.usage.usage_count"}}</th>
                         <th>{{i18n "discourse_ai.usage.total_tokens"}}</th>
+                        <th>{{i18n "discourse_ai.usage.total_spending"}}</th>
+
                       </tr>
                     </thead>
                     <tbody>
@@ -478,6 +522,13 @@ export default class AiUsage extends Component {
                             class="ai-usage__models-cell"
                             title={{model.total_tokens}}
                           >{{number model.total_tokens}}</td>
+                          <td>
+                            {{this.totalSpending
+                              model.input_spending
+                              model.cached_input_spending
+                              model.output_spending
+                            }}
+                          </td>
                         </tr>
                       {{/each}}
                     </tbody>
@@ -511,6 +562,7 @@ export default class AiUsage extends Component {
                           }}</th>
                         <th>{{i18n "discourse_ai.usage.usage_count"}}</th>
                         <th>{{i18n "discourse_ai.usage.total_tokens"}}</th>
+                        <th>{{i18n "discourse_ai.usage.total_spending"}}</th>
                       </tr>
                     </thead>
                     <tbody>
@@ -535,6 +587,13 @@ export default class AiUsage extends Component {
                             class="ai-usage__users-cell"
                             title={{user.total_tokens}}
                           >{{number user.total_tokens}}</td>
+                          <td>
+                            {{this.totalSpending
+                              user.input_spending
+                              user.cached_input_spending
+                              user.output_spending
+                            }}
+                          </td>
                         </tr>
                       {{/each}}
                     </tbody>
diff --git a/config/locales/client.en.yml b/config/locales/client.en.yml
index f50dda7f..1811de61 100644
--- a/config/locales/client.en.yml
+++ b/config/locales/client.en.yml
@@ -239,6 +239,7 @@ en:
         net_request_tokens: "Net request tokens"
         cached_tokens: "Cached tokens"
         cached_request_tokens: "Cached request tokens"
+        total_spending: "Estimated cost"
         no_users: "No user usage data found"
         no_models: "No model usage data found"
         no_features: "No feature usage data found"
@@ -249,6 +250,7 @@ en:
           request_tokens: "Tokens used when the LLM tries to understand what you are saying"
           response_tokens: "Tokens used when the LLM responds to your prompt"
           cached_tokens: "Previously processed request tokens that the LLM reuses to optimize performance and cost"
+          total_spending: "Cumulative cost of all tokens used by the LLMs based on specified cost metrics added to LLM configuration settings"
         periods:
           last_day: "Last 24 hours"
           last_week: "Last week"
@@ -404,6 +406,10 @@ en:
         enabled_chat_bot: "Allow AI bot selector"
         vision_enabled: "Vision enabled"
         ai_bot_user: "AI bot User"
+        cost_input: "Input cost"
+        cost_cached_input: "Cached input cost"
+        cost_output: "Output cost"
+
         save: "Save"
         edit: "Edit"
         saved: "LLM model saved"
@@ -487,6 +493,10 @@ en:
           name: "We include this in the API call to specify which model we'll use"
           vision_enabled: "If enabled, the AI will attempt to understand images. It depends on the model being used supporting vision. Supported by latest models from Anthropic, Google, and OpenAI."
           enabled_chat_bot: "If enabled, users can select this model when creating PMs with the AI bot"
+          cost_input: "The input cost per 1M tokens for this model"
+          cost_cached_input: "The cached input cost per 1M tokens for this model"
+          cost_output: "The output cost per 1M tokens for this model"
+          cost_measure: "$/1M tokens"
         providers:
           aws_bedrock: "AWS Bedrock"
           anthropic: "Anthropic"
diff --git a/db/migrate/20250416215039_add_cost_metrics_to_llm_model.rb b/db/migrate/20250416215039_add_cost_metrics_to_llm_model.rb
new file mode 100644
index 00000000..1df3ccac
--- /dev/null
+++ b/db/migrate/20250416215039_add_cost_metrics_to_llm_model.rb
@@ -0,0 +1,9 @@
+# frozen_string_literal: true
+
+class AddCostMetricsToLlmModel < ActiveRecord::Migration[7.2]
+  def change
+    add_column :llm_models, :input_cost, :float
+    add_column :llm_models, :cached_input_cost, :float
+    add_column :llm_models, :output_cost, :float
+  end
+end
diff --git a/lib/completions/llm.rb b/lib/completions/llm.rb
index a9b67c83..5f82667e 100644
--- a/lib/completions/llm.rb
+++ b/lib/completions/llm.rb
@@ -30,9 +30,26 @@ module DiscourseAi
                       name: "claude-3-7-sonnet",
                       tokens: 200_000,
                       display_name: "Claude 3.7 Sonnet",
+                      input_cost: 3,
+                      cached_input_cost: 0.30,
+                      output_cost: 15,
+                    },
+                    {
+                      name: "claude-3-5-haiku",
+                      tokens: 200_000,
+                      display_name: "Claude 3.5 Haiku",
+                      input_cost: 0.80,
+                      cached_input_cost: 0.08,
+                      output_cost: 4,
+                    },
+                    {
+                      name: "claude-3-opus",
+                      tokens: 200_000,
+                      display_name: "Claude 3 Opus",
+                      input_cost: 15,
+                      cached_input_cost: 1.50,
+                      output_cost: 75,
                     },
-                    { name: "claude-3-5-haiku", tokens: 200_000, display_name: "Claude 3.5 Haiku" },
-                    { name: "claude-3-opus", tokens: 200_000, display_name: "Claude 3 Opus" },
                   ],
                   tokenizer: DiscourseAi::Tokenizer::AnthropicTokenizer,
                   endpoint: "https://api.anthropic.com/v1/messages",
@@ -61,6 +78,8 @@ module DiscourseAi
                       endpoint:
                         "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash-lite",
                       display_name: "Gemini 2.0 Flash Lite",
+                      input_cost: 0.075,
+                      output_cost: 0.30,
                     },
                   ],
                   tokenizer: DiscourseAi::Tokenizer::GeminiTokenizer,
@@ -69,11 +88,46 @@ module DiscourseAi
                 {
                   id: "open_ai",
                   models: [
-                    { name: "o3-mini", tokens: 200_000, display_name: "o3 Mini" },
-                    { name: "o1", tokens: 200_000, display_name: "o1" },
-                    { name: "gpt-4.1", tokens: 800_000, display_name: "GPT-4.1" },
-                    { name: "gpt-4.1-mini", tokens: 800_000, display_name: "GPT-4.1 Mini" },
-                    { name: "gpt-4.1-nano", tokens: 800_000, display_name: "GPT-4.1 Nano" },
+                    {
+                      name: "o3-mini",
+                      tokens: 200_000,
+                      display_name: "o3 Mini",
+                      input_cost: 1.10,
+                      cached_input_cost: 0.55,
+                      output_cost: 4.40,
+                    },
+                    {
+                      name: "o1",
+                      tokens: 200_000,
+                      display_name: "o1",
+                      input_cost: 15,
+                      cached_input_cost: 7.50,
+                      output_cost: 60,
+                    },
+                    {
+                      name: "gpt-4.1",
+                      tokens: 800_000,
+                      display_name: "GPT-4.1",
+                      input_cost: 2,
+                      cached_input_cost: 0.5,
+                      output_cost: 8,
+                    },
+                    {
+                      name: "gpt-4.1-mini",
+                      tokens: 800_000,
+                      display_name: "GPT-4.1 Mini",
+                      input_cost: 0.40,
+                      cached_input_cost: 0.10,
+                      output_cost: 1.60,
+                    },
+                    {
+                      name: "gpt-4.1-nano",
+                      tokens: 800_000,
+                      display_name: "GPT-4.1 Nano",
+                      input_cost: 0.10,
+                      cached_input_cost: 0.025,
+                      output_cost: 0.40,
+                    },
                   ],
                   tokenizer: DiscourseAi::Tokenizer::OpenAiTokenizer,
                   endpoint: "https://api.openai.com/v1/chat/completions",
@@ -86,11 +140,15 @@ module DiscourseAi
                       name: "Meta-Llama-3.3-70B-Instruct",
                       tokens: 131_072,
                       display_name: "Llama 3.3 70B",
+                      input_cost: 0.60,
+                      output_cost: 1.20,
                     },
                     {
                       name: "Meta-Llama-3.1-8B-Instruct",
                       tokens: 16_384,
                       display_name: "Llama 3.1 8B",
+                      input_cost: 0.1,
+                      output_cost: 0.20,
                     },
                   ],
                   tokenizer: DiscourseAi::Tokenizer::Llama3Tokenizer,
diff --git a/lib/completions/report.rb b/lib/completions/report.rb
index 6982b9a7..7290a6d3 100644
--- a/lib/completions/report.rb
+++ b/lib/completions/report.rb
@@ -33,6 +33,27 @@ module DiscourseAi
         stats.total_requests || 0
       end
 
+      def total_spending
+        total = total_input_spending + total_output_spending + total_cached_input_spending
+        total.round(2)
+      end
+
+      def total_input_spending
+        model_costs.sum { |row| row.input_cost.to_f * row.total_request_tokens.to_i / 1_000_000.0 }
+      end
+
+      def total_output_spending
+        model_costs.sum do |row|
+          row.output_cost.to_f * row.total_response_tokens.to_i / 1_000_000.0
+        end
+      end
+
+      def total_cached_input_spending
+        model_costs.sum do |row|
+          row.cached_input_cost.to_f * row.total_cached_tokens.to_i / 1_000_000.0
+        end
+      end
+
       def stats
         @stats ||=
           base_query.select(
@@ -46,6 +67,24 @@ module DiscourseAi
           ]
       end
 
+      def model_costs
+        @model_costs ||=
+          base_query
+            .joins("LEFT JOIN llm_models ON llm_models.name = language_model")
+            .group(
+              "llm_models.name, llm_models.input_cost, llm_models.output_cost, llm_models.cached_input_cost",
+            )
+            .select(
+              "llm_models.name",
+              "llm_models.input_cost",
+              "llm_models.output_cost",
+              "llm_models.cached_input_cost",
+              "SUM(COALESCE(request_tokens, 0)) as total_request_tokens",
+              "SUM(COALESCE(response_tokens, 0)) as total_response_tokens",
+              "SUM(COALESCE(cached_tokens, 0)) as total_cached_tokens",
+            )
+      end
+
       def guess_period(period = nil)
         period = nil if %i[day month hour].include?(period)
         period ||
@@ -76,7 +115,15 @@ module DiscourseAi
       def user_breakdown
         base_query
           .joins(:user)
-          .group(:user_id, "users.username", "users.uploaded_avatar_id")
+          .joins("LEFT JOIN llm_models ON llm_models.name = language_model")
+          .group(
+            :user_id,
+            "users.username",
+            "users.uploaded_avatar_id",
+            "llm_models.input_cost",
+            "llm_models.output_cost",
+            "llm_models.cached_input_cost",
+          )
           .order("usage_count DESC")
           .limit(USER_LIMIT)
           .select(
@@ -87,12 +134,21 @@ module DiscourseAi
             "SUM(COALESCE(cached_tokens,0)) as total_cached_tokens",
             "SUM(COALESCE(request_tokens,0)) as total_request_tokens",
             "SUM(COALESCE(response_tokens,0)) as total_response_tokens",
+            "SUM(COALESCE(request_tokens, 0)) * COALESCE(llm_models.input_cost, 0) / 1000000.0 as input_spending",
+            "SUM(COALESCE(response_tokens, 0)) * COALESCE(llm_models.output_cost, 0) / 1000000.0 as output_spending",
+            "SUM(COALESCE(cached_tokens, 0)) * COALESCE(llm_models.cached_input_cost, 0) / 1000000.0 as cached_input_spending",
           )
       end
 
       def feature_breakdown
         base_query
-          .group(:feature_name)
+          .joins("LEFT JOIN llm_models ON llm_models.name = language_model")
+          .group(
+            :feature_name,
+            "llm_models.input_cost",
+            "llm_models.output_cost",
+            "llm_models.cached_input_cost",
+          )
           .order("usage_count DESC")
           .select(
             "case when coalesce(feature_name, '') = '' then '#{UNKNOWN_FEATURE}' else feature_name end as feature_name",
@@ -101,12 +157,21 @@ module DiscourseAi
             "SUM(COALESCE(cached_tokens,0)) as total_cached_tokens",
             "SUM(COALESCE(request_tokens,0)) as total_request_tokens",
             "SUM(COALESCE(response_tokens,0)) as total_response_tokens",
+            "SUM(COALESCE(request_tokens, 0)) * COALESCE(llm_models.input_cost, 0) / 1000000.0 as input_spending",
+            "SUM(COALESCE(response_tokens, 0)) * COALESCE(llm_models.output_cost, 0) / 1000000.0 as output_spending",
+            "SUM(COALESCE(cached_tokens, 0)) * COALESCE(llm_models.cached_input_cost, 0) / 1000000.0 as cached_input_spending",
           )
       end
 
       def model_breakdown
         base_query
-          .group(:language_model)
+          .joins("LEFT JOIN llm_models ON llm_models.name = language_model")
+          .group(
+            :language_model,
+            "llm_models.input_cost",
+            "llm_models.output_cost",
+            "llm_models.cached_input_cost",
+          )
           .order("usage_count DESC")
           .select(
             "language_model as llm",
@@ -115,6 +180,9 @@ module DiscourseAi
             "SUM(COALESCE(cached_tokens,0)) as total_cached_tokens",
             "SUM(COALESCE(request_tokens,0)) as total_request_tokens",
             "SUM(COALESCE(response_tokens,0)) as total_response_tokens",
+            "SUM(COALESCE(request_tokens, 0)) * COALESCE(llm_models.input_cost, 0) / 1000000.0 as input_spending",
+            "SUM(COALESCE(response_tokens, 0)) * COALESCE(llm_models.output_cost, 0) / 1000000.0 as output_spending",
+            "SUM(COALESCE(cached_tokens, 0)) * COALESCE(llm_models.cached_input_cost, 0) / 1000000.0 as cached_input_spending",
           )
       end
 
diff --git a/spec/fabricators/llm_model_fabricator.rb b/spec/fabricators/llm_model_fabricator.rb
index 87b1ef25..d61bb915 100644
--- a/spec/fabricators/llm_model_fabricator.rb
+++ b/spec/fabricators/llm_model_fabricator.rb
@@ -8,6 +8,9 @@ Fabricator(:llm_model) do
   api_key "123"
   url "https://api.openai.com/v1/chat/completions"
   max_prompt_tokens 131_072
+  input_cost 10
+  cached_input_cost 2.5
+  output_cost 40
 end
 
 Fabricator(:anthropic_model, from: :llm_model) do
diff --git a/spec/requests/admin/ai_usage_controller_spec.rb b/spec/requests/admin/ai_usage_controller_spec.rb
index eae9285b..f411ed41 100644
--- a/spec/requests/admin/ai_usage_controller_spec.rb
+++ b/spec/requests/admin/ai_usage_controller_spec.rb
@@ -5,6 +5,7 @@ require "rails_helper"
 RSpec.describe DiscourseAi::Admin::AiUsageController do
   fab!(:admin)
   fab!(:user)
+  fab!(:llm_model)
   let(:usage_report_path) { "/admin/plugins/discourse-ai/ai-usage-report.json" }
 
   before { SiteSetting.discourse_ai_enabled = true }
@@ -35,6 +36,18 @@ RSpec.describe DiscourseAi::Admin::AiUsageController do
         )
       end
 
+      fab!(:log3) do
+        AiApiAuditLog.create!(
+          provider_id: 1,
+          feature_name: "ai_helper",
+          language_model: llm_model.name,
+          request_tokens: 300,
+          response_tokens: 150,
+          cached_tokens: 50,
+          created_at: 3.days.ago,
+        )
+      end
+
       it "returns correct data structure" do
         get usage_report_path
 
@@ -55,7 +68,7 @@ RSpec.describe DiscourseAi::Admin::AiUsageController do
             }
 
         json = response.parsed_body
-        expect(json["summary"]["total_tokens"]).to eq(450) # sum of all tokens
+        expect(json["summary"]["total_tokens"]).to eq(900) # sum of all tokens
       end
 
       it "filters by feature" do
@@ -79,6 +92,26 @@ RSpec.describe DiscourseAi::Admin::AiUsageController do
         expect(models.first["total_tokens"]).to eq(300)
       end
 
+      it "shows an estimated cost" do
+        get usage_report_path, params: { model: llm_model.name }
+
+        json = response.parsed_body
+        summary = json["summary"]
+        feature = json["features"].find { |f| f["feature_name"] == "ai_helper" }
+
+        expected_input_spending = llm_model.input_cost * log3.request_tokens / 1_000_000.0
+        expected_cached_input_spending =
+          llm_model.cached_input_cost * log3.cached_tokens / 1_000_000.0
+        expected_output_spending = llm_model.output_cost * log3.response_tokens / 1_000_000.0
+        expected_total_spending =
+          expected_input_spending + expected_cached_input_spending + expected_output_spending
+
+        expect(feature["input_spending"].to_s).to eq(expected_input_spending.to_s)
+        expect(feature["output_spending"].to_s).to eq(expected_output_spending.to_s)
+        expect(feature["cached_input_spending"].to_s).to eq(expected_cached_input_spending.to_s)
+        expect(summary["total_spending"].to_s).to eq(expected_total_spending.round(2).to_s)
+      end
+
       it "handles different period groupings" do
         get usage_report_path, params: { period: "hour" }
         expect(response.status).to eq(200)