撰写文档，AI 在一旁与你一起编辑。

OpenAI's flagship GPT-6 model for complex reasoning, coding, research, and long-running agent work, with vision and a 1M-token context window.

GPT 6 Astra in OpenAI's pro reasoning mode, which works longer on each answer for the highest quality on hard tasks.

OpenAI's life sciences model, with specialized biological reasoning and scientific tool use. Only organizations approved through OpenAI's trusted-access program can call it, for internal research.

OpenAI's cybersecurity model for investigating vulnerabilities, reproducing them in an authorized test environment and verifying fixes. Only organizations approved through OpenAI's Daybreak program can call it.

OpenAI's update to GPT 6 Sol, the cost-efficient high-end GPT-6 tier, for agentic coding and document work, with vision and a 1M-token context window.

OpenAI's fast, efficient GPT-6.1 tier for visual reasoning, computer use and clear writing, improving on GPT 5.6 Terra with a 1M-token context window.

GPT 6.1 Sol in OpenAI's pro reasoning mode, which works longer on each answer for higher quality on hard tasks.

GPT 6.1 Terra in OpenAI's pro reasoning mode, which works longer on each answer for higher quality on hard tasks.

OpenAI's cost-efficient high-end GPT-6 tier for demanding coding and agentic work, with vision and a 1M-token context window.

GPT 6 Sol in OpenAI's pro reasoning mode, which works longer on each answer for higher quality on hard tasks.

OpenAI's fast, low-cost GPT-6 tier for high-volume chat, classification, and light agentic work, with vision.

GPT 6 Luna in OpenAI's pro reasoning mode, which works longer on each answer for higher quality on hard tasks.

OpenAI's frontier model with advanced reasoning and broad multimodal capabilities.

OpenAI's fastest, most affordable GPT-5.6 tier — for high-volume, latency-sensitive, and budget-conscious workloads.

OpenAI's flagship GPT-5.6 tier — leads the family on every benchmark, for frontier coding, long-horizon agentic work, and research.

Anthropic's Sonnet-class model for everyday coding and agentic work, a direct upgrade to Sonnet 5 with vision, tool use and a 1M-token context window.

Anthropic's most agentic Sonnet — frontier-level coding and tool use at Sonnet speed and price, with vision.

Moonshot's flagship open-weight multimodal reasoning model, post-trained from Kimi K3 for long-horizon coding and agent work, with a 1M-token context.

Moonshot's 2.8T-parameter open-weight mixture-of-experts reasoning model (104B active) with a 1M-token context, built for long-horizon agent work, coding and research.

Moonshot's frontier open-weight model (1T parameters) built for long-horizon agentic coding, with a 262K context and native vision.

Moondream's vision-language model (9B total, 2B active parameters) for questions about images, captions, reading text in images and visual reasoning.

Anthropic's Opus model for long-running agentic coding and knowledge work, with always-on adaptive thinking, vision, and a 1M-token context window.

Anthropic's Fable model for the most demanding reasoning and long-horizon agentic work, improving on Fable 5.1 in coding, reasoning and vision at a lower price, with always-on adaptive thinking and a 1M-token context window.

Anthropic's Fable model for the most demanding reasoning and long-horizon agentic work, with always-on adaptive thinking, vision, and a 1M-token context window.

Anthropic's Claude 5-family model for the most demanding reasoning, coding, and agentic work — with vision and a 128K output budget.

Anthropic's most capable model — frontier reasoning, coding, and agentic tool use with vision.

Anthropic's fastest model, built for high-volume, latency-sensitive work such as classification, extraction, routing and subagent tasks, with vision, tool use and a 1M-token context window.

Anthropic's fast, cost-efficient model with strong reasoning, coding, and tool use at low latency.

Alibaba's flagship Qwen3.8 reasoning model for coding, agentic workflows and document analysis, with image input and a 1M-token context window.

A higher-throughput variant of Alibaba's Qwen3.8 Max, served as a separate tier at a higher price.

Alibaba's fast, low-cost multimodal Qwen3.8 reasoning model for coding assistance, agentic workflows and visual understanding, with a 1M-token context window.

Alibaba's omni-modal Qwen3.8 reasoning model built around agentic work, which reads text, images, audio and video.

Meta's frontier reasoning model for complex personal and professional agentic tasks, with image input, tool calling and a 1M-token context window.

Meta's multimodal reasoning model for long-running agentic, multi-agent and coding workflows, with image input and a 1M-token context window.

Meta's open-weight multimodal reasoning model (816B total, 42B active parameters) built for agentic coding, with image input and a 1M-token context window.

Meta's smallest and most efficient Muse model for the work agents repeat all day: tool calls, coding subtasks, routing, extraction and long document parsing, with image, video and audio input and a 1M-token context window.

xAI's flagship Grok model for coding, agentic tasks, and knowledge work, with vision and a 500K-token context window.

xAI's frontier Grok model with strong reasoning, coding, and agentic tool use, plus vision and real-time knowledge from X.

Google's flagship reasoning model for complex software engineering, deep mathematical and scientific problem solving and autonomous multi-step agents, with a shared 1M-token context.

Google's most capable Flash model, with clear gains over 3.7 Flash in software engineering, agentic tasks, and multi-step reasoning, plus vision and a 1M-token context window.

Google's most efficient multimodal Flash model: an updated reasoning stack with stronger coding and computer-use quality, at the same speed and scale as the rest of the Flash line.

Google's most capable multimodal model with deep reasoning and 1M token context.

Google's 26B mixture-of-experts model with long context and strong multimodal reasoning.

Google's flagship 31B dense model with 256K context and top-tier text and image understanding.

Mistral's largest open-weight model, a multimodal mixture-of-experts built for reasoning, coding and agent work, with vision, tool calling and a 512K-token context window.

Liquid AI's mixture-of-experts reasoning model with 24B total and 2.45B active parameters. It always reasons before answering and is built for agents that call tools.

NVIDIA's open multimodal mixture-of-experts model (120B total, 12B active parameters) for coding, reasoning, tool use and document understanding, with image input and a 1M-token context window.

Reflection's open-weight reasoning model for coding and long-horizon agentic work: a sparse mixture of experts with 501B total and 23B active parameters, a 1M-token context and tool calling.

Thinking Machines' open-weight multimodal mixture-of-experts model (975B total, 41B active parameters) for reasoning, coding and agent work, reading text, images and audio.

The smaller Inkling (276B total, 12B active parameters), close to the larger model on many benchmarks at lower cost and latency, reading text, images and audio.

MiniMax's low-cost coding and agent model with a 1M-token context and tunable thinking depth, on Runware and OpenRouter.

DeepSeek's fast mixture-of-experts model and the first on its Causal Encoder-Decoder architecture, with image input, a 1M-token context and native tool calling.

DeepSeek's high-capability V4 model (MIT) with a 1M-token context, dual thinking modes, and stronger agentic performance than V4 Flash.

The high-speed variant of Z.ai's GLM-5.3, with the same capabilities at 1.5 to 2 times the output throughput and a higher price.

Z.ai's natively multimodal GLM-5.3 tier for efficient coding and long-horizon agent tasks, with image input and tool calling.

Z.ai's flagship open-weights coding model, with a 1M-token context and stronger long-horizon agentic work than GLM-5.2.

Z.ai's flagship long-horizon coding model, with a 1M-token context and deep-thinking reasoning for end-to-end engineering work.

ByteDance's fast Seed 2.1 reasoning model with a 256K context and output window, native tool calling and image input.

ByteDance's flagship Seed 2.0 reasoning model for complex analysis, coding and agentic tool use, with image input and a 256K context.

Balanced Seed 2.0 reasoning model with tool calling and image input at a lower price than Pro.

The smallest, cheapest Seed 2.0 model for fast everyday tasks, with tool calling and image input.

Preview of ByteDance's coding-tuned Seed 2.0 model for code generation and agentic programming, with tool calling.

Google's efficient 2B multimodal model supporting text, image, and audio input.

Google's efficient 4B multimodal model — stronger reasoning than E2B with text, image, and audio.

Google's 12B multimodal Gemma 4 model with strong text and image understanding.

Google's medical multimodal model fine-tuned for healthcare and biomedical tasks.

Large medical variant for complex clinical reasoning and biomedical analysis.

Mistral's 24B instruction-tuned model with vision input, robust function calling, and a 128K context.

Mistral's 14B dense edge model (Apache 2.0) tuned for fast on-device reasoning and tool use.

Mistral's efficient 8B dense edge model (Apache 2.0) for on-device chat, reasoning, and tool use.

Mistral's compact 3B dense edge model (Apache 2.0) for fast on-device text generation and tool use.

OpenAI's open-weight 20B reasoning model (Apache 2.0) with configurable reasoning effort and native tool calling — runs on 16 GB.

Microsoft's 14B state-of-the-art open model with strong reasoning, math, and code.

Microsoft's compact 3.8B model with a 128K context and native function calling.

Microsoft's 14B reasoning model fine-tuned for chain-of-thought on math, science, and code.

Ai2's fully open 32B model — open weights, data, and training recipe — with long chain-of-thought reasoning.

Ai2's fully open 7B model with open weights, training data, and recipe.

Ai2's fully open 32B model with open weights, training data, and recipe.

Poolside's 33B (3B active) MoE model for agentic coding and long-horizon local software work, with native reasoning and tool use.

Poolside's 33B (3B active) MoE agentic-coding model — Q8 quantization for higher fidelity.

Poolside's 33B (3B active) MoE agentic-coding model — full-precision BF16 weights.

Alibaba's ultra-compact 0.6B Qwen3 model — runs in-process via HuggingFace or through Ollama, with tool use.

Meta's 1B instruction-tuned model — the smallest Llama, runnable in-process via HuggingFace or through Ollama.

Google's compact 1B Gemma 3 model — runs in-process via HuggingFace or through Ollama.

Liquid's 24B foundation model with efficient long-context reasoning.

Meta's 8B instruction-tuned model with 128K context and strong general reasoning.

Meta's compact 3B instruction-tuned model — fast on-device text generation.

IBM's compact 3B instruction-tuned model with multilingual support, tool use, and structured JSON output.

IBM's 8B dense decoder-only foundation model for general instruction-following, RAG, and code tasks.

IBM's flagship 30B dense foundation model — strong reasoning across business, coding, and multilingual tasks.

IBM's compact 3B instruction-tuned model with multilingual support, tool use, and structured JSON output.

IBM's 8B dense foundation model for general instruction-following, RAG, and code tasks, with native tool use.

IBM's flagship 30B foundation model — strong reasoning across business, coding, and multilingual tasks.

Meta Superintelligence Labs' 30B multimodal model for always-on local agents — tuned for tool use, long tasks, and failure recovery.

NVIDIA's 30B mixture-of-experts model with 3B active parameters, built as the execution layer for always-on agents.

NVIDIA's compact vision-language model (74B total, 9B active parameters), distilled from Nemotron 3.5 Super and tuned to fit a 24 GB NVIDIA GPU for coding, reasoning and tool use.

NVIDIA's small vision-language model (8B parameters, 4B effective), post-trained from Gemma 4 E4B for fast local conversations, image understanding and multi-turn tool use; a bundled draft model speeds up generation.

Prism ML's 9B vision model trained for ternary weights (-1, 0 or +1), about half the size of a standard 4-bit model at close to full-precision quality, with tool calling, thinking and a 256K context.

Prism ML's 27B vision model trained for ternary weights (-1, 0 or +1), about half the size of a standard 4-bit model at close to full-precision quality, with tool calling, thinking and a 256K context.

NVIDIA's multimodal Nemotron mixture-of-experts model unifying text and image understanding for enterprise Q&A, summarization, and document intelligence.

Alibaba's 27B multimodal Qwen 3.8 model with a 256K context — substantial gains in coding, research, and long-horizon agentic tasks, with native image and video understanding.

Alibaba's 27B multimodal Qwen 3.6 model with a 256K context and strong vision-language reasoning.

Alibaba's flagship 35B multimodal Qwen 3.6 model for top-tier vision-language reasoning.

Alibaba's compact 2B multimodal Qwen model with vision input and a 256K context — fast on-device text+image reasoning.

Alibaba's efficient 4B multimodal Qwen model with vision input and strong multilingual reasoning.

Alibaba's 9B multimodal Qwen model balancing vision-language reasoning with on-device performance.

Liquid's efficient 8B Liquid Foundation Model tuned for long-context reasoning on-device.

Liquid AI's first omni-modal LFM3 model (2.2B parameters), built for on-device use; the app sends it text and images.

Cohere's compact code-specialized model tuned for fast local code generation, completion, and refactoring.

Deep Reinforce's 9B self-improving agentic-coding model (MIT) with 256K context and native tool use.

Deep Reinforce's flagship 35B self-improving agentic-coding model (MIT) with 256K context — state-of-the-art open-source coding for its size.

Deep Reinforce's 9B self-improving model (MIT) with 256K context, image input, and native tool use, tuned for reasoning, coding, and agentic work.

Deep Reinforce's 35B MoE self-improving model (MIT) with 256K context, image input, and native tool use, strong across reasoning, coding, and agentic tasks.

Deep Reinforce's 122B MoE self-improving model (MIT) with 256K context, image input, and native tool use, the largest Ornith 1.5 size that runs on a single high-memory machine.

DeepSeek's compact 1.5B reasoning model distilled from Qwen2.5 — chain-of-thought on-device.

DeepSeek's 8B reasoning model distilled from Qwen3 (R1-0528) with strong math, code, and logic.

DeepSeek's 14B reasoning model distilled from Qwen2.5 with exceptional math and coding benchmarks.

DeepSeek's 32B reasoning model distilled from Qwen2.5 for top-tier chain-of-thought reasoning.

Alibaba's compact 4B instruction model with strong reasoning and multilingual capabilities.