# OpenRouter 推出 Deep Research 工具及多款新模型

- 来源：OpenRouter：Announcements（RSS）
- 作者：OpenRouter
- 发布时间：2025-03-06 08:00
- AIHOT 分数：56
- AIHOT 标记：精选
- AIHOT 链接：https://aihot.virxact.com/items/cmq29zxqy00i2slophtn5ro4j
- 原文链接：https://openrouter.ai/blog/deep-research-and-many-new-models

## 精选理由

第一个通过 API 可调用的 deep research 工具，直接返回带引用结果，想在自己产品里加 research 能力的开发者不容错过。

## AI 摘要

OpenRouter API 上线首个深度研究工具，API 响应中直接包含完整引用。同时平台新增多款模型。

## 正文

Deep Research & Many New Models — OpenRouter Blog

Deep Research & Many New Models

OpenRouter · 3/6/2025

On this page

New Sonar Models, Including Deep Research, Now Available:

More New Models on OpenRouter:

New Sonar Models, Including Deep Research, Now Available:

Sonar Deep Research: Engineered explicitly for comprehensive research tasks, Sonar Deep Research automates in-depth searches, iteratively evaluates sources, and synthesizes structured analyses for complex subjects across multiple domains.

Sonar Reasoning Pro: Advanced reasoning capabilities powered by DeepSeek R1 with Chain of Thought (CoT). Perfect for sophisticated queries requiring detailed analysis and extensive sourcing.

Sonar Pro: Enhanced model capable of handling complex, multi-step queries with a larger context window, ideal for detailed searches and nuanced follow-ups.

Note: Pricing for these models includes additional charges based on Perplexity’s internal search and reasoning steps. Full pricing details are available here.

More New Models on OpenRouter:

Creative & Interactive Storytelling:

Skyfall 36B v2: Fine-tuned from Mistral Small for enhanced creativity, nuanced storytelling, and immersive roleplay.

Wayfarer Large 70B: Optimized for realistic, narrative-rich interactive fiction with meaningful consequences.

Anubis Pro 105B v1: Expanded variant of Llama 3.3 70B with deeper character portrayals, emotional intelligence, and coherent storytelling.

Advanced & Efficient:

Reka Flash 3: A 21B-parameter model optimized for efficient general chat, coding, reasoning, and function-calling tasks with explicit internal reasoning tags.

Cohere Command A: 111B parameters with a 256K context window for multilingual, coding, and agentic tasks.

Jamba 1.6 Large: 94B active parameters, multilingual proficiency, structured JSON outputs, and tool-use capabilities.

Jamba 1.6 Mini: Compact 12B parameter model, multilingual, structured outputs, and efficient tool use.

Multimodal Gemma 3 Models (free!):

Gemma 3 27B: The largest of the Gemma 3 family!

Gemma 3 12B: Fast & Efficient.

Gemma 3 4B: Faster than 12B, suitable for small local deployments.

Gemma 3 1B: Compact, ultra-fast, non-multimodal Gemma.

Try them out now across the chatroom and the API