adds claude-opus model family

.env.example

@@ -40,9 +40,10 @@ NODE_ENV=production
 
 # Which model types users are allowed to access.
 # The following model families are recognized:
-# turbo | gpt4 | gpt4-32k | gpt4-turbo | dall-e | claude | gemini-pro | mistral-tiny | mistral-small | mistral-medium | mistral-large | aws-claude | azure-turbo | azure-gpt4 | azure-gpt4-32k | azure-gpt4-turbo
-# By default, all models are allowed except for 'dall-e'. To allow DALL-E image
-# generation, uncomment the line below and add 'dall-e' to the list.
+# turbo | gpt4 | gpt4-32k | gpt4-turbo | dall-e | claude | claude-opus | gemini-pro | mistral-tiny | mistral-small | mistral-medium | mistral-large | aws-claude | azure-turbo | azure-gpt4 | azure-gpt4-32k | azure-gpt4-turbo
+# By default, all models are allowed except for 'dall-e' and 'claude-opus'. To
+# allow DALL-E image generation, uncomment the line below and add 'dall-e' to
+# the list.
 # ALLOWED_MODEL_FAMILIES=turbo,gpt4,gpt4-32k,gpt4-turbo,claude,gemini-pro,mistral-tiny,mistral-small,mistral-medium,mistral-large,aws-claude,azure-turbo,azure-gpt4,azure-gpt4-32k,azure-gpt4-turbo
 
 # URLs from which requests will be blocked.

src/info-page.ts

@@ -17,7 +17,8 @@ const MODEL_FAMILY_FRIENDLY_NAME: { [f in ModelFamily]: string } = {
   "gpt4-32k": "GPT-4 32k",
   "gpt4-turbo": "GPT-4 Turbo",
   "dall-e": "DALL-E",
-  "claude": "Claude",
+  "claude": "Claude (Sonnet)",
+  "claude-opus": "Claude (Opus)",
   "gemini-pro": "Gemini Pro",
   "mistral-tiny": "Mistral 7B",
   "mistral-small": "Mixtral Small", // Originally 8x7B, but that now refers to the older open-weight version. Mixtral Small is a newer closed-weight update to the 8x7B model.

src/proxy/middleware/request/onproxyreq/add-anthropic-preamble.ts

@@ -7,11 +7,12 @@ import { HPMRequestCallback } from "../index";
  * know this without trying to send the request and seeing if it fails. If a
  * key is marked as requiring a preamble, it will be added here.
  */
-export const addAnthropicPreamble: HPMRequestCallback = (
-  _proxyReq,
-  req
-) => {
-  if (!isTextGenerationRequest(req) || req.key?.service !== "anthropic") {
+export const addAnthropicPreamble: HPMRequestCallback = (_proxyReq, req) => {
+  if (
+    !isTextGenerationRequest(req) ||
+    req.key?.service !== "anthropic" ||
+    req.outboundApi !== "anthropic-text"
+  ) {
     return;
   }
 

src/proxy/middleware/request/preprocessors/validate-context-size.ts

@@ -69,10 +69,14 @@ export const validateContextSize: RequestPreprocessor = async (req) => {
     modelMax = 100000;
   } else if (model.match(/^claude-2/)) {
     modelMax = 200000;
+  } else if (model.match(/^claude-3/)) {
+    modelMax = 200000;
   } else if (model.match(/^gemini-\d{3}$/)) {
     modelMax = GOOGLE_AI_MAX_CONTEXT;
   } else if (model.match(/^mistral-(tiny|small|medium)$/)) {
     modelMax = MISTRAL_AI_MAX_CONTENT;
+  } else if (model.match(/^anthropic\.claude-3-sonnet/)) {
+    modelMax = 200000;
   } else if (model.match(/^anthropic\.claude-v2:\d/)) {
     modelMax = 200000;
   } else if (model.match(/^anthropic\.claude/)) {

src/shared/key-management/anthropic/provider.ts

@@ -13,8 +13,8 @@ export type AnthropicModel =
   | "claude-v1-100k"
   | "claude-2"
   | "claude-2.1"
-  | "claude-3-opus-20240229"
-  | "claude-3-sonnet-20240229";
+  | "claude-3-opus-20240229" // new expensive model
+  | "claude-3-sonnet-20240229" // new cheap claude2 sidegrade
 
 export type AnthropicKeyUpdate = Omit<
   Partial<AnthropicKey>,
@@ -90,7 +90,7 @@ export class AnthropicKeyProvider implements KeyProvider<AnthropicKey> {
       const newKey: AnthropicKey = {
         key,
         service: this.service,
-        modelFamilies: ["claude"],
+        modelFamilies: ["claude", "claude-opus"],
         isDisabled: false,
         isOverQuota: false,
         isRevoked: false,
@@ -107,6 +107,7 @@ export class AnthropicKeyProvider implements KeyProvider<AnthropicKey> {
           .slice(0, 8)}`,
         lastChecked: 0,
         claudeTokens: 0,
+        "claude-opusTokens": 0,
       };
       this.keys.push(newKey);
     }

src/shared/models.ts

@@ -22,7 +22,7 @@ export type OpenAIModelFamily =
   | "gpt4-32k"
   | "gpt4-turbo"
   | "dall-e";
-export type AnthropicModelFamily = "claude";
+export type AnthropicModelFamily = "claude" | "claude-opus";
 export type GoogleAIModelFamily = "gemini-pro";
 export type MistralAIModelFamily =
   | "mistral-tiny"
@@ -51,6 +51,7 @@ export const MODEL_FAMILIES = (<A extends readonly ModelFamily[]>(
   "gpt4-turbo",
   "dall-e",
   "claude",
+  "claude-opus",
   "gemini-pro",
   "mistral-tiny",
   "mistral-small",
@@ -96,6 +97,7 @@ export const MODEL_FAMILY_SERVICE: {
   "gpt4-32k": "openai",
   "dall-e": "openai",
   claude: "anthropic",
+  "claude-opus": "anthropic",
   "aws-claude": "aws",
   "azure-turbo": "azure",
   "azure-gpt4": "azure",
@@ -122,6 +124,7 @@ export function getOpenAIModelFamily(
 
 export function getClaudeModelFamily(model: string): ModelFamily {
   if (model.startsWith("anthropic.")) return getAwsBedrockModelFamily(model);
+  if (model.includes("opus")) return "claude-opus";
   return "claude";
 }
 
@@ -146,7 +149,8 @@ export function getMistralAIModelFamily(model: string): MistralAIModelFamily {
   }
 }
 
-export function getAwsBedrockModelFamily(_model: string): ModelFamily {
+export function getAwsBedrockModelFamily(model: string): ModelFamily {
+  if (model.includes("opus")) return "claude-opus";
   return "aws-claude";
 }
 

src/shared/stats.ts

@@ -29,6 +29,9 @@ export function getTokenCostUsd(model: ModelFamily, tokens: number) {
     case "claude":
       cost = 0.000008;
       break;
+    case "claude-opus":
+      cost = 0.000015;
+      break;
     case "mistral-tiny":
       cost = 0.00000031;
       break;

src/shared/users/user-store.ts

@@ -34,6 +34,7 @@ const INITIAL_TOKENS: Required<UserTokenCounts> = {
   "gpt4-turbo": 0,
   "dall-e": 0,
   claude: 0,
+  "claude-opus": 0,
   "gemini-pro": 0,
   "mistral-tiny": 0,
   "mistral-small": 0,