🧰AI tool library

included 444 AI models · 22 manufacturers · Continuously updated

共 444 个模型
🧠 Model AI CodingAI Writing Thinking Machines Lab

Inkling 2 Mini

Thinking Machines 开源的循环 Transformer,3.4B 激活/26.4B 总参,为本地部署设计
📊 3.4B 激活 / 26.4B 总参(M 💰 开源可自建;云端输入 0 📐 256K(部分 provide
🧠 Model AI WritingAI Coding OpenAI

GPT-6.2 Sol

GPT-6.2 系列日常工作马,105 万上下文,性价比向,适合日常写作与代码
📊 Not publicly disclosed 💰 输入 2 美元 / 输出 📐 1,050,000 token
🧠 Model AI WritingAI Coding OpenAI

GPT-6.2 Astra

GPT-6.2 系列旗舰,105 万上下文,面向真实工程、专业工作与科学研究
📊 Not publicly disclosed 💰 输入 8 美元 / 输出 📐 1,050,000 token
🧠 Model AI CodingAI Writing Google

Gemini 4 Neon

Gemini 4 系列快中价档,1M 上下文、支持视频/音频/图像输入,主打仓库级编程与 Agent
📊 Not publicly disclosed 💰 输入 3 美元 / 输出 📐 1,048,576 token
🧠 Model AI WritingAI Coding Google

Gemini 4 Hydrogen

Google 新一代主力模型,替掉 3.8 Flash,四档思考、速度快 2.5 倍、Terminal-Bench 三倍
📊 Not publicly disclosed 💰 API 1.5美元/百万 📐 1M token
🧠 Model AI Coding Meta

Muse Ember

Meta’s Lightest Agent Execution Model: $0.10 per Million Input Tokens, 1M Context Window, Multimodal Input Support
📊 Not publicly disclosed 💰 $0.10 per Million Input Tokens 📐 1,048,576 token
🧠 Model AI Coding Anthropic

Claude Fable 5.2

Anthropic flagship, 1M context, 128K output, input cost reduced to $8 per million tokens, zero data retention supported
📊 Not publicly disclosed 💰 $8 per million tokens input 📐 1M-token input
🧠 Model AI WritingAI Coding Liquid AI

LFM3-1.7B

1.73B on-device text model, 256K context window, 4-bit quantization occupies only ~1GB RAM, fine-tuned for Agent tool invocation and structured output
📊 1.73B, triple-hybrid architecture: 13-layer gated short RNN 💰 $0.08 per 100 tokens (input) 📐 256K tokens (input + output)
🧠 Model AI Video Google

Gemini Omni 1.2 Flash

Google’s native multimodal video generation model: 15-second clips, conversational multi-turn editing, $0.07 per second at 720p
📊 Closed-source cloud models 💰 $0.07/second (720p) 📐 131K token
🧠 Model AI Image Gen OpenAI

GPT-Image-2.6

OpenAI’s image model, fully launched in November 2026, generates images in 6 seconds, natively supports GIF, and exports SVG; its API offers two variants: Flare and Sunburst
📊 Not disclosed (API-side: Flare is the speed-optimized variant) 💰 Free within ChatGPT 📐 Not applicable (image generation model)
🧠 Model AI WritingAI Coding Baidu (Wenxin)

Wenxin 5.5 Flash Base

Baidu’s 280-billion-parameter MoE base model, open-sourced on November 5, 2026; activates 14 billion parameters per token and supports 1M-context inference
📊 280B total parameters / 14B activated (M 💰 Weights open-sourced and free; self-hosted inference enabled 📐 Pre-trained on 128K context; fine-tuning and inference supported
🧠 Model AI CodingAI Productivity Tencent (Hunyuan)

Hunyuan Hy4

Tencent’s 770B MoE Flagship Release Version: 49B Activated, 1M Context Length, Image Input Support, $0.417 per Million Input Tokens
📊 770B Total Parameters / 49B Activated (MoE) 💰 $0.417 per Million Input Tokens 📐 1M Tokens Context Length, Maximum Output Length
🧠 Model AI Coding Baseten

joey-1

OpenAI quietly increased quotas for paying users. According to
Google integrated file editing capabilities into Gemini Google delivered results that left Wall Street underwhelmed Computer science rests on a foundational concept called ETH (Expo
🧠 Model AI Audio NVIDIA

Nemotron 3.5 Transcribe

34B open-weight speech model supporting transcription in 40 languages and handling 1-hour audio; deployable on a single H100 GPU
📊 34B total parameters / 4.3B activated MoE 💰 Open-weight and free; requires your own GPU 📐 65,536 tokens maximum context length
🧠 Model AI Coding Particle AI

Particle-1

An agent programming model post-trained on DeepSeek V4.1 Flash, with 1M context window and input pricing at $0.25 per million tokens
📊 DeepSeek V4.1 Flash 💰 $0.25 per million input tokens, $ 📐 1M tokens, maximum output 1
🧠 Model AI WritingAI Coding OpenAI

GPT-6.1 Terra

OpenAI’s mid-tier flagship model, excelling in visual reasoning and computer operation, with a 1.05M context window and priced at half that of GPT-6.1 Sol
📊 Unreleased (closed-source) 💰 $1 / $5 📐 1.05M tokens, per request
🧠 Model AI CodingAI Writing Dark Side of the Moon (Kimi)

Kimi K3.1 Pro

Moonshot 2.8T Total Parameters / 104B Activated Flagship, 1M Context Window, Agent and Coding Capabilities Approaching Top Closed-Source Tier
📊 2.8T Total Parameters / 104B Activated (M 💰 $2 / $10 per Million 📐 1M Tokens, Max Output Approx.
🧠 Model AI Image GenAI Design Canva

Canva Craft

Canva’s first in-house image generation model, trained on design assets, supports 2K resolution and 30 aspect ratios, and outputs layered design files
📊 Not publicly disclosed 💰 Free for Canva users 📐 Unpublished
🧠 Model AI Coding Recursive

Recursive Flint

A 40B open-source MoE programming model with only 2.5B activated parameters; free to download and run offline on laptops; SWE-Bench verified outstanding performance
📊 40B MoE, 2.5B activated 💰 Free (Apache 2. 📐 Unpublished
🧠 Model AI Coding NVIDIA

Nemotron 3.5 Super VL

NVIDIA’s 120B open-source Agent model, designed for enterprise private deployment, matches flagship 550B models in instruction following and tool-calling accuracy
📊 120B parameters, plus Bolt 74B-A9 💰 Open-source and free; users must provide their own GPU for deployment 📐 Not publicly disclosed; designed for long-horizon Agent workflows
🧠 Model AI Image Gen Elorian

Elorian Glimpse

A Visual Reasoning Model That Thinks While It Sees—Scientific Chart and Video Understanding on Par with GPT-6 Astra, Priced at Just 6% of Astra’s Cost
📊 Not Publicly Disclosed, Post-Training on Open-Source Weight Base 💰 $0.60 per Input / $3 per Output 📐 Unpublished
🧠 Model AI CodingAI Writing Meta

Muse Spark Open

Meta’s Most Powerful Open-Source Multimodal Model in the U.S.—816B Parameters, 1M-Token Context Window, Designed Specifically for Coding and Agent Tasks, Runs Locally on Mac
📊 816 Billion Parameters (MoE; Activated Parameter Count Not Disclosed) 💰 Weights Open-Sourced and Free; Local Execution Requires 📐 1 Million Tokens
🧠 Model AI Writing Google

Gemini 4 Argon

Google’s most advanced frontier model, ranked #1 on the Vals Index, excelling at deep reasoning for programming, finance, law, cybersecurity, and scientific research
📊 Not publicly disclosed 💰 API: $2 per million input tokens 📐 Long context (exact length not disclosed)
🧠 Model AI Writing Meta

Muse Wonder

Meta’s first frontier model, benchmarked against GPT-6 Astra and Fable 5.1, featuring active memory and Muse agents
📊 Unreleased (frontier model) 💰 Input / million tokens 📐 Unpublished
🧠 Model AI Coding Microsoft

Microsoft-Decision-1

Microsoft Decision Model—for routing, classification, verification, and other structured decision-making tasks—35x faster than GPT-6 Sol, with free output
📊 9B (Qwen3.5-9B Post-Training) 💰 Input bash.042/100 📐 Unpublished
🧠 Model AI WritingAI Coding Anthropic

Claude Haiku 5.5

Anthropic’s fastest and most cost-effective small model: $0.10 per million input tokens—75% cheaper than its predecessor, with performance leaps of several times, ideal for high-concurrency summarization, classification, and coding subagents
📊 Unreleased (Haiku-class small model) 💰 $0.10 per hundred input tokens 📐 200K tokens
🧠 Model AI WritingAI Coding OpenAI

GPT-6.1 Sol

OpenAI efficiency model—five times cheaper than GPT-6 Astra, strong performance in agent coding and professional tasks
📊 Not publicly disclosed 💰 ~GPT-6 Astra 📐 Unpublished
🧠 Model AI WritingAI Coding Anthropic

Claude Sonnet 5.5

Anthropic’s mid-tier flagship model: 30% faster, 30% cheaper, 1M context window, reasoning always enabled, excels at coding and document tasks
📊 Not publicly disclosed 💰 $0 per million tokens for input/output 📐 1M tokens
🧠 Model AI ProductivityAI Coding Cohere

Cohere Command A+

Hearing Enhancement
Meta calls this “FDA-cleared software for mild to moderate hearing loss,” launching this year as part of the Meta One subscription or as a standalone add-on for $149. 💰 bash.30/.50 416 models total
🧠 Model AI WritingAI Coding Anthropic

Claude Opus 5.5

Anthropic security-hardened model: 85% reduction in boundary violations, 40% lower runtime cost, aligned with Fable 5.1
📊 Not publicly disclosed 💰 Lower runtime cost than Opus 5 📐 Unpublished
Scroll to Top
Tool Picks
📬 Weekly AI Picks