Claude Opus 5.5

ModelProviderOverall scoreCore strengthsContextScore snapshot

1Claude Opus 5.5 Anthropic 93.5 Agentic coding · Computer use · 1M-token context 1,000,000 tokens C94R100S100M84A90O94 → 2Claude Sonnet 5.5 Anthropic 93.0 Terminal coding · Computer use · 1M-token context 1,000,000 tokens C100R90S100M84A91O93 → 3GPT-6 Astra OpenAI 90.6 1.05M-token context and 128K output · State-of-the-art reported coding, science, and computer-use results · Broad Responses API tool support 1,050,000 tokens C92R90S84M100A87O91 → 4GPT-6.1 Sol OpenAI 88.6 Complex coding · Computer use · Lower cached-input price 1,050,000 tokens C88R87S85M100A83O89 → 5Gemini 4 Argon Google 88.2 Long software-engineering workflows · Video understanding · 1M-token maximum output Not disclosed C89R93S87M71A100O88 → 6Claude Fable 5.1 Anthropic 86.3 1M-token context · 128K maximum output · Strong coding and professional-work benchmark coverage 1,000,000 tokens C81R97S94M84A76O86 → 7GPT-6 Sol OpenAI 77.8 Text and image input · Configurable reasoning · 1.05M-token context 1,050,000 tokens C69R79S82M81A80O78 → 8Qwen3.8-Max-0902 Alibaba Cloud 73.6 Source-backed model profile 1,000,000 tokens C61R71S91M74A72O74 → 9Grok 4.7 xAI 70.4 Long-running agents · Visual and professional work 500,000 tokens C41R71S92M65A85O70 → 10GLM-5.3 Z.ai 67.8 Open-weight model · Long-horizon coding 1,000,000 tokens C66R69S90M36A80O68 → 11GLM-5.3-Flash Z.ai 66.3 Open weights under MIT · 1M-token context and 128K maximum output · Native text, image, video, and file input 1,048,576 tokens C52R66S89M48A77O66 → 12Grok 4.6 xAI 66.2 Source-backed model profile 500,000 tokens C33R71S87M55A86O66 → 12Gemini 3.8 Flash Google 66.2 1.05M-token context and 64K output · Configurable low, medium, and high thinking · Strong reported coding and agentic results 1,050,000 tokens C31R79S77M68A77O66 → 14Kimi K3 Moonshot AI 65.2 Source-backed model profile 1,000,000 tokens C20R77S83M71A74O65 → 15DeepSeek-V4.1-Flash DeepSeek 64.6 1M-token context with native text, image, and video input · Causal Encoder-Decoder (CED) architecture with Engram memory for high inference speed · State-of-the-art reported coding and reasoning performance (GPQA Diamond: 90.9%, Terminal-Bench 2.1: 90.6%) 1,000,000 tokens C42R64S87M42A89O65 → 16Qwen3.8-Max Alibaba Cloud 62.9 Native visual input · 1M-token context · Long-horizon coding 1,000,000 tokens C30R71S87M65A63O63 → 17GPT-6 Luna OpenAI 60.8 Text and image input · Configurable reasoning · 1.05M-token context 1,050,000 tokens C20R64S78M74A68O61 → 18MiniMax M3 MiniMax 38.7 Native image and video input · 1M-token context · Open weights 1,000,000 tokens C3R64S68M32A27O39 → 19Mistral Medium 3.5 Mistral 16.2 Open weights · Visual input · Coding and agentic tasks 256,000 tokens C0R23S41M10A8O16 → —Qwen3.8-Flash-Next Alibaba Cloud — Source-backed model profile 262,144 tokens C—R—S—M—A—O— → —Claude Sonnet 5 Anthropic — Source-backed model profile Not disclosed C—R—S—M—A—O— → —Claude Haiku 4.5 Anthropic — Low latency · Lower token cost 200,000 tokens C—R—S—M—A—O— → —Claude Mythos 5.1 Anthropic — Source-backed model profile Not disclosed C—R—S—M—A—O— → —Gemini 3.8 Flash Cyber Google — Source-backed model profile Not disclosed C—R—S—M—A—O— → —GPT-5.6 Sol OpenAI — Source-backed model profile Not disclosed C—R—S—M—A—O— → —GPT-5.6 Terra OpenAI — Source-backed model profile Not disclosed C—R—S—M—A—O— → —GPT-5.6 Luna OpenAI — Source-backed model profile Not disclosed C—R—S—M—A—O— → —Gemini 3.5 Flash Google — Source-backed model profile Not disclosed C—R—S—M—A—O— → —Gemini 3.6 Flash Google — Source-backed model profile Not disclosed C—R—S—M—A—O— → —Claude Fable 5 Anthropic — Source-backed model profile Not disclosed C—R—S—M—A—O— → —Claude Opus 5 Anthropic — Source-backed model profile Not disclosed C—R—S—M—A—O— → —Claude Opus 4.8 Anthropic — Source-backed model profile Not disclosed C—R—S—M—A—O— → —Claude Opus 4.7 Anthropic — Source-backed model profile Not disclosed C—R—S—M—A—O— → —DeepSeek V4 Pro DeepSeek — Source-backed model profile 1,000,000 tokens C—R—S—M—A—O— → —DeepSeek V4 Flash DeepSeek — Source-backed model profile 1,000,000 tokens C—R—S—M—A—O— → —OpenAI GPT-5.4 OpenAI — Strong SWE-Bench Pro result · High GPQA Diamond result · Broad multimodal and tool-use coverage Not disclosed C—R—S—M—A—O— → —OpenAI GPT-5.2 Thinking OpenAI — Strong GPQA and FrontierMath performance · Useful cross-category benchmark coverage · Explicit API model naming Not disclosed C—R—S—M—A—O— → —Claude Opus 4.6 Anthropic — Strong official SWE-bench Verified result · Agentic workflow positioning · Long-context workflow emphasis Not disclosed C—R—S—M—A—O— → —Gemini 3.1 Pro Google — High GPQA Diamond and ARC-AGI-2 results · Strong multimodal and agentic benchmark coverage · 1M input token context Not disclosed C—R—S—M—A—O— → —Gemini 3 Pro Google — Strong GPQA Diamond and MMMU-Pro scores · Broad multimodal model coverage Not disclosed C—R—S—M—A—O— → —DeepSeek-R1 DeepSeek — Strong open reasoning model positioning · Reported GPQA, MMLU-Pro, FRAMES, and math benchmark results Not disclosed C—R—S—M—A—O— → —Gemini 3.7 Flash Google — Broad native multimodal input · 1M-token context · Configurable low, medium, and high reasoning 1,048,576 tokens C—R—S—M—A—O— → —Qwen3.8-Flash Alibaba Cloud — 1M-token production context · 125B total and 6B activated language-model parameters · Strong coding, agentic, and multimodal benchmark coverage 1,000,000 tokens C—R—S—M—A—O— → —DeepSeek-V3 DeepSeek — Open-weight architecture rivaling leading closed frontier models · Exceptional cost efficiency via Multi-head Latent Attention and FP8 execution · Strong reported mathematics and coding performance (MATH-500: 90.2%, HumanEval-Mul: 82.6%) 131,072 tokens C—R—S—M—A—O— → —Tencent Hunyuan-Large Tencent — Open-source MoE model with 52B activated parameters per token · 256K long context with KV cache compression (CLA/GQA) · Strong academic and code performance outperforming Llama 3.1-70B 262,144 tokens C—R—S—M—A—O— → —Hunyuan-T1 Tencent — Hybrid Transformer-Mamba MoE architecture achieving 60-80 tokens/sec decoding speed · Post-training focused 96.7% compute on RL for complex logic and competition mathematics · High verified scores on MMLU-Pro (87.2) and GPQA-Diamond (69.3) 28,672 tokens C—R—S—M—A—O— →

Overall averages coding, reasoning, research, multimodal and agentic indices on a 0–100 scale. Original benchmark scores remain in the evidence; missing axes produce no overall rank.