最佳视觉模型

Inkling

Thinking Machines Lab's open-weights general-purpose multimodal Mixture-of-Experts model with 975B total parameters and 41B active parameters. Inkling accepts text, image, and audio inputs, produces text, and is designed for agentic and tool-use systems, coding assistants, chatbots, and retrieval-augmented generation.

Kimi K3

Moonshot AI's flagship Kimi model for frontier intelligence, agentic coding, knowledge work, and deep reasoning. Kimi K3 supports a 1-million-token context window for long-running software engineering and research workflows.

GPT-5.6 Luna

The fast, low-cost tier of OpenAI's GPT-5.6 series, optimized for high-volume, latency-sensitive tasks such as classification, extraction, routing, and lightweight agentic steps. Approaches the larger GPT-5.6 tiers on many benchmarks while running several times faster at a fraction of the price.

Muse Spark 1.1

Meta Superintelligence Labs' updated flagship, building on Muse Spark with stronger agentic reasoning, more reliable multi-agent orchestration, and improved multimodal understanding across voice, text, and image. Extends the context window and reduces latency and reasoning token usage while raising coding and tool-use accuracy. Powers Meta AI across its product ecosystem.

Gemini 3.5 Pro

Google DeepMind's flagship Gemini model, rebuilt on a new foundation with a 2M-token context window and a Deep Think reasoning mode for the hardest math, coding, and multimodal tasks. Natively multimodal across text, image, audio, and video with streaming function calling and strong long-context grounding.

GPT-5.6 Sol

OpenAI's flagship model in the GPT-5.6 series, advancing coding, scientific reasoning, long-horizon planning, and agentic workflows while improving reliability and efficiency on demanding real-world tasks. Adds a max reasoning-effort setting and an ultra mode that spawns subagents for complex, multi-step work.

GPT-5.6 Terra

The balanced tier of OpenAI's GPT-5.6 series, trading a small amount of peak quality for markedly lower latency and cost. Retains strong reasoning, coding, and agentic tool use with configurable reasoning effort, making it a default choice for production workloads that need frontier capability at scale.

Grok 4.5

xAI's strongest model to date, built to excel at coding, agentic tasks, and knowledge work and co-developed alongside coding tools for real-world software engineering. Features real-time information access, extended reasoning, and large-context tool use with an OpenAI-compatible API.

Claude Sonnet 5

Anthropic's most capable Sonnet-class model, bringing frontier coding, agentic, and professional-work performance to the midsize tier while closing the gap with Opus 4.8 at a lower price. Supports adaptive thinking with selectable reasoning effort levels, a 1M-token context window, and text, image, and file inputs. Codenamed Fennec.

Claude Fable 5

Anthropic's first publicly available Mythos-class model, exceeding the capabilities of any model the company has previously made generally available. State-of-the-art on nearly all tested benchmarks, with exceptional performance in software engineering, knowledge work, vision, and scientific research. Its lead grows on longer and more complex tasks. Ships with built-in safeguards that route sensitive cybersecurity, biology, chemistry, and distillation queries to Claude Opus 4.8.

Claude Mythos 5

Anthropic's frontier Mythos-class model — the same underlying model as Claude Fable 5 but with safeguards lifted in some areas. It has the strongest cybersecurity capabilities of any model in the world, alongside state-of-the-art performance in software engineering, knowledge work, vision, and scientific research. Access is restricted to a small group of trusted cyberdefenders and infrastructure providers through Project Glasswing.

Gemma 4 12B

Google's medium-size open-weight model with 12 billion parameters from the Gemma 4 family. Encoder-free unified multimodal architecture that natively processes text, image, audio, and video inputs without dedicated encoders. Features a 256K context window and supports 140+ languages. First medium-sized model capable of natively ingesting audio. Suitable for local deployment on GPUs with 16GB VRAM.

MiniMax M3

MiniMax's frontier open-weight model with 1M-token context window, native multimodality (text, image, video), and strong coding capabilities. Built on MiniMax Sparse Attention (MSA) architecture, achieving 59% on SWE-Bench Pro with significantly improved efficiency at long context.

Claude Opus 4.8

Anthropic's most advanced model, building on Opus 4.7 with improvements across benchmarks in coding, agentic skills, reasoning, and knowledge work. Features enhanced honesty, better tool use efficiency, dynamic workflows support, and improved alignment.

Qwen 3.7 Plus

Alibaba's multimodal variant in the Qwen 3.7 family, optimized for vision understanding and multimodal tasks. Ranked

Gemini 3.1 Flash-Lite

Google's most cost-efficient Gemini model optimized for high-volume, low-latency use cases. Delivers 2.5x faster time to first token versus Gemini 2.5 Flash with full multimodal support. Ideal for agentic tasks, data extraction, translation, and classification.

Gemini 3.5 Flash

Google DeepMind's balanced Gemini 3.5 model that pairs Pro-line reasoning quality with Flash-line latency and cost. Natively multimodal across text, image, audio, and video with a 1M-token context window, configurable thinking levels, and streaming function calling, tuned for high-throughput production workloads.

Gemini 3 Flash

Google's balanced model combining Gemini 3 Pro's reasoning capabilities with the Flash line's latency, efficiency, and cost. Features configurable thinking levels, multimodal function responses, and streaming function calling for complex agentic workflows.

MiniCPM-V 4.6

Ultra-efficient multimodal language model from OpenBMB built on SigLIP2-400M and Qwen3.5-0.8B (~1B parameters). Supports single-image, multi-image, and video understanding with mixed 4x/16x visual token compression. Designed for edge deployment on iOS, Android, and HarmonyOS.

GPT-5.5

OpenAI's most capable model designed for complex real-world work including coding, online research, information analysis, and document creation. Features advanced agentic capabilities with tool search and multi-step task execution.

Qwen 3.6 27B

Alibaba's dense 27B parameter model that outperforms its own 397B MoE predecessor on agentic coding benchmarks. Strong multilingual and reasoning capabilities released under Apache 2.0.

Qwen 3.6 35B-A3B

Alibaba's efficient Mixture-of-Experts model with 35B total parameters and 3B active per token. Frontier-level agentic coding performance with 73.4% on SWE-bench Verified and 92.7 on AIME 2026. Released under Apache 2.0.

GPT-5.4 Mini

OpenAI's compact reasoning model optimized for coding, computer use, and subagent tasks. Approaches GPT-5.4 performance on several benchmarks while running more than 2x faster.

Muse Spark

Meta Superintelligence Labs' first model, featuring advanced reasoning, multimodal understanding, and agentic capabilities. Processes voice, text, and image inputs with tool use and multi-agent orchestration. Powers Meta AI across its product ecosystem.

Qwen 3.6 Plus

Alibaba's proprietary flagship model in the Qwen 3.6 family, targeting enterprise AI workflows with stronger agentic coding capability, visual coding support, and end-to-end enterprise engineering features.

Gemma 4 31B

Google's flagship open-weight dense model with 31B parameters. All parameters active per forward pass. Ranks among top open models with strong performance on AIME 2026 (89.2%) and MMLU Pro (85.2%). Supports vision and extended context.

Gemma 4 26B

Google's high-performance open-weight dense model with 26 billion parameters from the Gemma 4 family. Supports multimodal inputs including text and images with a 256K extended context window. Strong reasoning and code generation capabilities with all parameters active per forward pass.

Gemma 4 31B

Google's flagship open-weight dense model with 31 billion parameters from the Gemma 4 family. All parameters active per forward pass with top-tier performance on reasoning benchmarks including AIME 2026 and MMLU Pro. Supports vision and extended 256K context window.

Claude Opus 4.7

Anthropic's latest and most advanced model with state-of-the-art reasoning, coding, and analysis capabilities. Features improved tool use, extended thinking, and enhanced safety alignment.

Grok 4.3

xAI's latest and most intelligent model with strong agentic tool calling, minimal hallucinations, and configurable reasoning. Supports 1M token context window with competitive pricing.

GPT-5.4

OpenAI's frontier reasoning model combining advances in coding, reasoning, and agentic workflows. Features 1.1M token context window and strong performance on complex multi-step problems.

GPT-5.5 Pro

OpenAI's premium tier model with extended reasoning capabilities, higher accuracy on complex tasks, and priority access. Optimized for professional and enterprise workloads requiring maximum quality.

Gemini 3.1 Pro

Google's latest flagship multimodal model with state-of-the-art performance on reasoning, coding, and multimodal understanding. Features native tool use, grounding, and million-token context window.

Grok 4.20

xAI's multi-agent capable model with 2M token context window. Available in reasoning, non-reasoning, and multi-agent variants for diverse enterprise workloads.

Grok 4

xAI's latest model with real-time information access, strong reasoning capabilities, and competitive performance on coding and analysis tasks. Features improved tool use and multimodal understanding.

Claude Opus 4.6

Anthropic's most capable model in the Claude 4 family, excelling at complex analysis, extended reasoning, scientific research, and advanced code generation. Features significantly improved accuracy and reduced hallucinations.

Claude Sonnet 4.6

Anthropic's balanced model offering strong performance at lower cost and latency than Opus. Excellent for everyday coding, analysis, and content generation tasks with good reasoning capabilities.

Mistral Large 3

Mistral AI's largest open-weight model with 41B active parameters (675B total MoE). State-of-the-art general-purpose multimodal model with 256K context window and powerful agentic capabilities. Released under Apache 2.0.

Grok 4.1 Fast

xAI's fast and cost-effective model with 2M token context window. Offers both reasoning and non-reasoning modes at significantly lower pricing than flagship models.

Claude Haiku 4.5

Anthropic's fastest model with near-frontier intelligence. Optimized for high-throughput, low-latency applications requiring quick responses at minimal cost. Supports extended thinking.

Claude Sonnet 4.5

Anthropic's previous-generation balanced model with strong coding and analysis capabilities. Offers excellent price-performance ratio for production workloads requiring reliable quality.

Gemini 2.5 Pro

Google's high-capability reasoning model with adaptive thinking for complex agentic and multimodal challenges. Features 1M token context window and strong performance on coding and scientific tasks.

Gemini 2.5 Flash

Google's cost-effective model optimized for high throughput tasks. Balances speed and intelligence with strong multimodal capabilities and 1M token context window.

GPT-5

OpenAI's fifth-generation flagship model with significant improvements in reasoning, multimodal understanding, and code generation. Features enhanced instruction following and expanded context window.

Llama 4 Maverick

Meta's quality-focused MoE model with 17B active parameters (400B total, 128 experts). Targets quality-critical tasks with benchmark scores competitive with GPT-4o and Gemini 2.5 Pro.

Llama 4 Scout

Meta's efficient MoE model with 17B active parameters (109B total, 16 experts). Supports up to 10M token context — the longest of any production model. Strong performance on reasoning and multilingual tasks.

Gemma 3 12B

Google's mid-size open-weight model with 12 billion parameters from the Gemma 3 family. Supports multimodal inputs including text and images with a 128K context window. Strong performance on reasoning and code generation tasks at moderate compute cost.

Gemma 3 27B

Google's largest open-weight model in the Gemma 3 family with 27 billion parameters. Supports multimodal inputs including text and images with a 128K context window. Delivers strong performance across reasoning, code generation, and vision tasks, competitive with larger proprietary models.

Gemma 3 4B

Google's compact open-weight model with 4 billion parameters from the Gemma 3 family. Supports multimodal inputs including text and images with a 128K context window. Balances efficiency and capability for vision and language tasks.

Mistral Small 3.1

Mistral AI's Small 3.1 model with 24B parameters offering efficient multimodal capabilities including vision, function calling, and code generation with a large 128K context window.

Llama 3.2 11B Vision Instruct

Meta's multimodal open-weight model with 11 billion parameters from the Llama 3.2 family. Supports both text and image inputs, enabling visual understanding tasks alongside standard text generation. Suitable for applications requiring vision capabilities at moderate scale.

Llama 3.2 90B Vision Instruct

Meta's largest multimodal open-weight model with 90 billion parameters from the Llama 3.2 family. Delivers strong performance on both text and image understanding tasks with competitive results on visual reasoning benchmarks. Designed for high-quality inference requiring vision capabilities.

Claude 3 Opus

Anthropic's most powerful model in the Claude 3 family, excelling at complex analysis, nuanced content generation, scientific reasoning, and code generation with extended context support.

GPT-4

OpenAI's flagship large language model with advanced reasoning, instruction following, and code generation capabilities. Supports multimodal inputs including text and images.