@agentionai/agents 1.2.0 → 1.3.0

package/dist/agents/BaseAgent.d.ts

@@ -56,6 +56,14 @@ export type TokenUsage = {
      * Generation throughput: `output_tokens` over `generationMs`. Falls back to
      * `totalMs` when the first-token time is unknown (an unstreamed call), in
      * which case it is an end-to-end rate rather than a pure generation rate.
+     *
+     * Accurate wherever thinking is streamed, because the first thinking chunk
+     * starts the generation window — verified on Anthropic extended thinking and
+     * on DeepSeek via OpenRouter. It over-reports on OpenAI's Responses API,
+     * which does not stream raw reasoning: there the thinking finishes before the
+     * first visible token, so it lands inside `timeToFirstTokenMs` while its
+     * tokens still count toward `output_tokens`. For the visible-output rate on
+     * that path, divide `output_tokens - reasoning_tokens` by `generationMs`.
      */
     outputTokensPerSecond?: number;
 };

package/dist/agents/anthropic/ClaudeAgent.js

@@ -43,7 +43,7 @@ class ClaudeAgent extends BaseAgent_1.BaseAgent {
             ? { authToken: config.apiKey }
             : { apiKey: config.apiKey });
         this.config = {
-            model: config.model || "claude-3-5-haiku-latest",
+            model: config.model || "claude-haiku-4-5",
             maxTokens: config.maxTokens || 1024,
             disableParallelToolUse,
             metadata,

package/dist/agents/google/GeminiAgent.js

@@ -35,7 +35,7 @@ class GeminiAgent extends BaseAgent_1.BaseAgent {
         const responseMimeType = config.responseMimeType ?? vendorConfig.responseMimeType;
         const responseSchema = config.responseSchema ?? vendorConfig.responseSchema;
         this.config = {
-            model: config.model || "gemini-2.0-flash",
+            model: config.model || "gemini-flash-latest",
             maxTokens: config.maxTokens || 1024,
             apiKey: config.apiKey,
             temperature: config.temperature,

package/dist/agents/model-types.d.ts

@@ -8,19 +8,19 @@
  * You can also provide any custom string for newer models not yet listed.
  * @see https://docs.anthropic.com/en/docs/about-claude/models
  */
-export type ClaudeModel = "claude-opus-4-7" | "claude-opus-4-6" | "claude-sonnet-4-6" | "claude-opus-4-5" | "claude-sonnet-4-5" | "claude-haiku-4-5" | "claude-haiku-4-5-20251001" | (string & Record<never, never>);
+export type ClaudeModel = "claude-opus-5" | "claude-sonnet-5" | "claude-fable-5" | "claude-opus-4-8" | "claude-opus-4-7" | "claude-opus-4-6" | "claude-sonnet-4-6" | "claude-opus-4-5" | "claude-opus-4-5-20251101" | "claude-sonnet-4-5" | "claude-sonnet-4-5-20250929" | "claude-haiku-4-5" | "claude-haiku-4-5-20251001" | (string & Record<never, never>);
 /**
  * Supported Google Gemini models.
  * You can also provide any custom string for newer models not yet listed.
  * @see https://ai.google.dev/gemini-api/docs/models/gemini
  */
-export type GeminiModel = "gemini-flash-latest" | "gemini-flash-lite-latest" | "gemini-3.0-pro" | "gemini-3.0-flash" | "gemini-2.5-flash-lite" | "gemini-2.0-flash-exp" | "gemini-2.0-flash" | "gemini-2.0-flash-lite" | (string & {});
+export type GeminiModel = "gemini-pro-latest" | "gemini-flash-latest" | "gemini-flash-lite-latest" | "gemini-3.6-flash" | "gemini-3.5-flash" | "gemini-3.5-flash-lite" | "gemini-3.1-pro-preview" | "gemini-3.1-flash-lite" | "gemini-3.1-flash-lite-preview" | "gemini-3-pro-preview" | "gemini-3-flash-preview" | "gemini-2.5-pro" | "gemini-2.5-flash" | "gemini-2.5-flash-lite" | "gemini-2.0-flash" | "gemini-2.0-flash-001" | "gemini-2.0-flash-lite" | "gemini-2.0-flash-lite-001" | (string & {});
 /**
  * Supported Mistral models.
  * You can also provide any custom string for newer models not yet listed.
  * @see https://docs.mistral.ai/getting-started/models/
  */
-export type MistralModel = "mistral-large-latest" | "mistral-small-latest" | "ministral-8b-latest" | "ministral-8b-2410" | "ministral-3b-latest" | "ministral-3b-2410" | "codestral-latest" | "codestral-2405" | "mistral-embed" | "mistral-moderation-latest" | "mistral-moderation-2411" | (string & {});
+export type MistralModel = "mistral-large-latest" | "mistral-large-2512" | "mistral-medium-latest" | "mistral-medium-3.5" | "mistral-small-latest" | "mistral-small-2603" | "magistral-small-latest" | "ministral-14b-latest" | "ministral-8b-latest" | "ministral-8b-2512" | "ministral-3b-latest" | "ministral-3b-2512" | "codestral-latest" | "codestral-2508" | "devstral-latest" | "devstral-medium-latest" | "mistral-embed" | "codestral-embed" | "mistral-moderation-latest" | "mistral-moderation-2603" | "mistral-ocr-latest" | (string & {});
 /**
  * Popular Ollama models (locally hosted).
  * You can also provide any custom string for models you have pulled.

package/dist/agents/openai-compatible/OpenAICompatibleAgent.js

@@ -320,13 +320,17 @@ class OpenAICompatibleAgent extends BaseAgent_1.BaseAgent {
         let reasoningContent = "";
         const toolCallAcc = new Map();
         let finishReason = null;
+        let streamUsage;
         for await (const chunk of stream) {
-            // Final chunk carrying usage (choices is empty)
-            if (chunk.choices.length === 0) {
-                if (chunk.usage)
-                    this.accumulateStreamUsage(chunk.usage);
+            // Usage can ride on any chunk: OpenAI sends it on a final choice-less
+            // chunk, OpenRouter attaches it to the last content chunk (the one
+            // carrying finish_reason). Keep the most recent and fold it in once the
+            // stream ends — it is a running total for the turn, not a delta, so
+            // taking the last one covers both layouts without double-counting.
+            if (chunk.usage)
+                streamUsage = chunk.usage;
+            if (chunk.choices.length === 0)
                 continue;
-            }
             const choice = chunk.choices[0];
             finishReason = choice.finish_reason ?? finishReason;
             const delta = choice.delta;
@@ -366,6 +370,10 @@ class OpenAICompatibleAgent extends BaseAgent_1.BaseAgent {
                 }
             }
         }
+        // Before any early return below, so a turn that hits the token limit or
+        // continues into a tool call still reports what it spent.
+        if (streamUsage)
+            this.accumulateStreamUsage(streamUsage);
         if (finishReason === "length") {
             const error = new AgentError_1.MaxTokensExceededError("Response exceeded maximum token limit", this.config.maxTokens || 1024);
             this.emit(AgentEvent_1.AgentEvent.MAX_TOKENS_EXCEEDED, error);

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@agentionai/agents",
   "author": "Laurent Zuijdwijk",
-  "version": "1.2.0",
+  "version": "1.3.0",
   "description": "Agent Library",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",