{"ok":true,"source":"tensorfeed.ai","lastUpdated":"2026-10-19","benchmarks":[{"id":"swe_bench","name":"SWE-bench","description":"Real-world software engineering tasks from GitHub issues (SWE-bench Verified)","maxScore":100},{"id":"mmlu_pro","name":"MMLU-Pro","description":"General knowledge and reasoning across 57 subjects","maxScore":100},{"id":"human_eval","name":"HumanEval","description":"Python code generation and problem solving","maxScore":100},{"id":"gpqa_diamond","name":"GPQA Diamond","description":"Graduate-level science questions verified by domain experts","maxScore":100},{"id":"math","name":"MATH","description":"Competition-level mathematics problems","maxScore":100},{"id":"osworld_2","name":"OSWorld 2.0","description":"Computer use across real desktop applications and multi-step GUI tasks","maxScore":100},{"id":"browsecomp","name":"BrowseComp","description":"Agentic web search and browsing over hard-to-find facts","maxScore":100},{"id":"frontier_code","name":"FrontierCode v1.1","description":"Agentic coding on frontier software engineering tasks (Main split)","maxScore":100},{"id":"terminal_bench_4","name":"Terminal-Bench 4.0","description":"Long-horizon agentic work in a terminal across software, science, ML, operations, hardware, security, and media (66 tasks, all-or-nothing verifiers)","maxScore":100},{"id":"hle_tools","name":"Humanity's Last Exam (tools)","description":"Multidisciplinary expert-level reasoning with tool access","maxScore":100}],"models":[{"model":"Ember-1","provider":"Fireworks AI","released":"2026-09","scores":{"terminal_bench_4":20.71,"swe_bench":92.2},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, 2026-10-18 board update. The cell previously held 19.7."},"swe_bench":{"type":"vendor","by":"Fireworks AI","url":"https://fireworks.ai/blog/ember-1"}}},{"model":"Grok 4.7","provider":"xAI","released":"2026-09","scores":{"terminal_bench_4":28.79},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, xhigh effort."}}},{"model":"MiMo-V2.6-Pro","provider":"Xiaomi","released":"2026-09","scores":{"terminal_bench_4":31.31},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3."}}},{"model":"MiMo-V2.6-Flash","provider":"Xiaomi","released":"2026-09","scores":{"terminal_bench_4":24.24},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3."}}},{"model":"Claude Opus 5.5","provider":"Anthropic","released":"2026-09","scores":{"terminal_bench_4":65.15,"frontier_code":54.6,"hle_tools":67.7},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, max effort. 22 of 198 attempts were served by the Opus 5 or Opus 4.8 fallback; counting those as failures gives 58.08."},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode","note":"Main split, best reasoning mode (medium effort). Anthropic's own table reports 54.4 at max effort."},"hle_tools":{"type":"vendor","by":"Anthropic","url":"https://www.anthropic.com/claude-opus-5-5"}}},{"model":"Claude Sonnet 5.5","provider":"Anthropic","released":"2026-09","scores":{"terminal_bench_4":64.14,"frontier_code":52.1,"hle_tools":64.5},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, max effort. 4 of 198 attempts were served by the Sonnet 5 fallback; counting those as failures gives 62.63. Anthropic self-reports 70.6."},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode","note":"Main split, best reasoning mode (xhigh effort). Anthropic's own table reports the same 52.1 at xhigh and 46.2 at max effort."},"hle_tools":{"type":"vendor","by":"Anthropic","url":"https://www.anthropic.com/claude-sonnet-5-5"}}},{"model":"GPT-6.1 Sol","provider":"OpenAI","released":"2026-09","scores":{"terminal_bench_4":55.05,"frontier_code":50.2},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, max effort."},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode","note":"Main split, best reasoning mode (medium effort)."}}},{"model":"GPT-6 Sol","provider":"OpenAI","released":"2026-09","scores":{"terminal_bench_4":44.44,"frontier_code":49.3},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, max effort."},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode","note":"Main split, max effort."}}},{"model":"GPT-6 Luna","provider":"OpenAI","released":"2026-09","scores":{"terminal_bench_4":13.64,"frontier_code":42.4},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, max effort."},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode","note":"Main split, max effort."}}},{"model":"Ternary Bonsai 2 27B","provider":"PrismML","released":"2026-09","scores":{"swe_bench":60.8},"provenance":{"swe_bench":{"type":"vendor","by":"PrismML","url":"https://prismml.com/news/bonsai-2-27b"}}},{"model":"DeepSeek V4.1 Flash","provider":"DeepSeek","released":"2026-09","scores":{"terminal_bench_4":19.7,"gpqa_diamond":90.9,"hle_tools":63.9},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, high effort."},"gpqa_diamond":{"type":"vendor","by":"DeepSeek","url":"https://api-docs.deepseek.com/updates/"},"hle_tools":{"type":"vendor","by":"DeepSeek","url":"https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"}}},{"model":"Qwen3.8 2.4T-A95B","provider":"Alibaba","released":"2026-08","scores":{"gpqa_diamond":92.6,"hle_tools":56.2},"provenance":{"gpqa_diamond":{"type":"vendor","by":"Alibaba","url":"https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B"},"hle_tools":{"type":"vendor","by":"Alibaba","url":"https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B"}}},{"model":"Qwen3.8 27B","provider":"Alibaba","released":"2026-08","scores":{"swe_bench":86,"mmlu_pro":84.34,"gpqa_diamond":89.2,"osworld_2":48},"provenance":{"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"vendor","by":"Alibaba","url":"https://huggingface.co/Qwen/Qwen3.8-27B"},"osworld_2":{"type":"vendor","by":"Alibaba","url":"https://huggingface.co/Qwen/Qwen3.8-Flash-Next","note":"Partial score, from the comparison column of the Qwen3.8-Flash-Next card. Binary completion on the same run is 19.4."}}},{"model":"Muse Wonder","provider":"Meta","released":"2026-10","scores":{"terminal_bench_4":49.49},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, 2026-10-18 board update. Listed by Vals as Wonder, the API model. Muse Wonder Full, which is limited to Muse Code subscribers, scores 58.08 on the same board and is not tracked as a separate row."}}},{"model":"Muse Spark Open","provider":"Meta","released":"2026-10","scores":{"terminal_bench_4":6.57},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, 2026-10-18 board update."}}},{"model":"Muse Glimmer 30B","provider":"Meta","released":"2026-08","scores":{"swe_bench":76,"gpqa_diamond":83.5},"provenance":{"swe_bench":{"type":"vendor","by":"Meta","url":"https://huggingface.co/meta-models/Muse-Glimmer-30B"},"gpqa_diamond":{"type":"vendor","by":"Meta","url":"https://huggingface.co/meta-models/Muse-Glimmer-30B"}}},{"model":"Nemotron 3.5 Lightning","provider":"NVIDIA","released":"2026-08","scores":{"swe_bench":51.56,"mmlu_pro":81.94,"gpqa_diamond":75.44,"browsecomp":36.97},"provenance":{"swe_bench":{"type":"vendor","by":"NVIDIA","url":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16"},"mmlu_pro":{"type":"vendor","by":"NVIDIA","url":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16"},"gpqa_diamond":{"type":"vendor","by":"NVIDIA","url":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16"},"browsecomp":{"type":"vendor","by":"NVIDIA","url":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16"}}},{"model":"Gemini 3.7 Flash","provider":"Google","released":"2026-08","scores":{"terminal_bench_4":12.12,"swe_bench":80.8,"mmlu_pro":90.12,"gpqa_diamond":94.5,"osworld_2":47.9,"frontier_code":43.6},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, high effort."},"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"independent","by":"Artificial Analysis","url":"https://artificialanalysis.ai/evaluations/gpqa-diamond"},"osworld_2":{"type":"vendor","by":"Google DeepMind","url":"https://deepmind.google/models/model-cards/gemini-3-7-flash/"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"}}},{"model":"Grok 4.6","provider":"xAI","released":"2026-08","scores":{"terminal_bench_4":17.17,"swe_bench":95.6,"mmlu_pro":89.4,"gpqa_diamond":94.9,"frontier_code":48},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, high effort."},"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"independent","by":"Artificial Analysis","url":"https://artificialanalysis.ai/evaluations/gpqa-diamond"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"}}},{"model":"GLM-5.3","provider":"Z.ai","released":"2026-08","scores":{"terminal_bench_4":38.89,"swe_bench":95.4,"mmlu_pro":86.77,"gpqa_diamond":88.13,"frontier_code":40.1,"hle_tools":62.5},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, max effort."},"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/models/zai_glm-5.3"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/models/zai_glm-5.3"},"gpqa_diamond":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/models/zai_glm-5.3"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"},"hle_tools":{"type":"vendor","by":"Z.ai","url":"https://z.ai/blog/glm-5.3"}}},{"model":"Qwen3.8-Max","provider":"Alibaba","released":"2026-08","scores":{"terminal_bench_4":34.34,"swe_bench":85.6,"mmlu_pro":88.6,"gpqa_diamond":92.6,"hle_tools":56.2},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3. Nine refusals are scored as failures."},"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"vendor","by":"Alibaba","url":"https://qwen.ai/blog?id=qwen3.8"},"hle_tools":{"type":"vendor","by":"Alibaba","url":"https://qwen.ai/blog?id=qwen3.8"}}},{"model":"GPT-5.6 Sol","provider":"OpenAI","released":"2026-07","scores":{"terminal_bench_4":37.88,"swe_bench":96.2,"mmlu_pro":89.1,"gpqa_diamond":94.6,"osworld_2":62.6,"browsecomp":90.4,"frontier_code":47.5},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, max effort."},"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"vendor","by":"OpenAI","url":"https://openai.com/index/gpt-5-6/"},"osworld_2":{"type":"vendor","by":"OpenAI","url":"https://openai.com/index/gpt-5-6/"},"browsecomp":{"type":"vendor","by":"OpenAI","url":"https://openai.com/index/gpt-5-6/"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"}}},{"model":"GPT-5.6 Terra","provider":"OpenAI","released":"2026-07","scores":{"terminal_bench_4":22.73,"swe_bench":95.4,"mmlu_pro":86.66,"gpqa_diamond":92.9,"osworld_2":50.2,"browsecomp":87.5,"frontier_code":41.3},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, max effort."},"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"vendor","by":"OpenAI","url":"https://openai.com/index/gpt-5-6/"},"osworld_2":{"type":"vendor","by":"OpenAI","url":"https://openai.com/index/gpt-5-6/"},"browsecomp":{"type":"vendor","by":"OpenAI","url":"https://openai.com/index/gpt-5-6/"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"}}},{"model":"GPT-5.6 Luna","provider":"OpenAI","released":"2026-07","scores":{"terminal_bench_4":11.62,"swe_bench":93,"mmlu_pro":86.04,"gpqa_diamond":92.3,"osworld_2":45.6,"browsecomp":83.3,"frontier_code":39.8},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, max effort."},"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"vendor","by":"OpenAI","url":"https://openai.com/index/gpt-5-6/"},"osworld_2":{"type":"vendor","by":"OpenAI","url":"https://openai.com/index/gpt-5-6/"},"browsecomp":{"type":"vendor","by":"OpenAI","url":"https://openai.com/index/gpt-5-6/"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"}}},{"model":"Gemini 3.6 Flash","provider":"Google","released":"2026-07","scores":{"swe_bench":79.6,"mmlu_pro":89.28,"gpqa_diamond":93.43,"osworld_2":33.8,"frontier_code":34.4},"provenance":{"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/gpqa"},"osworld_2":{"type":"vendor","by":"Google DeepMind","url":"https://deepmind.google/models/model-cards/gemini-3-7-flash/"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"}}},{"model":"Grok 4.5","provider":"xAI","released":"2026-07","scores":{"terminal_bench_4":8.59,"swe_bench":86.6,"mmlu_pro":89.22,"gpqa_diamond":92.93,"frontier_code":42.4},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, high effort."},"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/gpqa"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"}}},{"model":"Inkling","provider":"Thinking Machines","released":"2026-07","scores":{"terminal_bench_4":0.51,"swe_bench":77.6,"mmlu_pro":86.3,"gpqa_diamond":87.2,"browsecomp":77.1,"frontier_code":14,"hle_tools":46},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3."},"swe_bench":{"type":"vendor","by":"Thinking Machines","url":"https://thinkingmachines.ai/model-card/inkling/"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"vendor","by":"Thinking Machines","url":"https://thinkingmachines.ai/model-card/inkling/"},"browsecomp":{"type":"vendor","by":"Thinking Machines","url":"https://thinkingmachines.ai/model-card/inkling/"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"},"hle_tools":{"type":"vendor","by":"Thinking Machines","url":"https://thinkingmachines.ai/model-card/inkling/"}}},{"model":"GLM-5.2","provider":"Z.ai","released":"2026-06","scores":{"swe_bench":82.8,"mmlu_pro":86.71,"gpqa_diamond":91.2,"frontier_code":24.5,"hle_tools":54.7},"provenance":{"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/models/zai_glm-5.2"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/models/zai_glm-5.2"},"gpqa_diamond":{"type":"vendor","by":"Z.ai","url":"https://huggingface.co/zai-org/GLM-5.2"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"},"hle_tools":{"type":"vendor","by":"Z.ai","url":"https://huggingface.co/zai-org/GLM-5.2"}}},{"model":"MiniMax M3","provider":"MiniMax","released":"2026-06","scores":{"terminal_bench_4":1.01,"swe_bench":80.5,"mmlu_pro":84.22,"gpqa_diamond":92.68,"browsecomp":83.5,"frontier_code":14.7},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3."},"swe_bench":{"type":"vendor","by":"MiniMax","url":"https://huggingface.co/MiniMaxAI/MiniMax-M3"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/models/minimax_MiniMax-M3"},"browsecomp":{"type":"vendor","by":"MiniMax","url":"https://huggingface.co/MiniMaxAI/MiniMax-M3"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"}}},{"model":"LongCat-2.0","provider":"Meituan","released":"2026-06","scores":{"gpqa_diamond":88.9,"browsecomp":79.9},"provenance":{"gpqa_diamond":{"type":"vendor","by":"Meituan","url":"https://huggingface.co/meituan-longcat/LongCat-2.0"},"browsecomp":{"type":"vendor","by":"Meituan","url":"https://huggingface.co/meituan-longcat/LongCat-2.0"}}},{"model":"GPT-5.5","provider":"OpenAI","released":"2026-04","scores":{"mmlu_pro":94.2,"human_eval":97.1,"gpqa_diamond":78.3,"math":95.8,"swe_bench":82.6,"frontier_code":43},"provenance":{"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"}}},{"model":"DeepSeek V4 Pro","provider":"DeepSeek","released":"2026-04","scores":{"mmlu_pro":91.5,"human_eval":94.8,"gpqa_diamond":92.4,"math":92.4,"swe_bench":80.6},"provenance":{"gpqa_diamond":{"type":"vendor","by":"DeepSeek","url":"https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"}}},{"model":"DeepSeek V4 Flash","provider":"DeepSeek","released":"2026-04","scores":{"mmlu_pro":85.2,"human_eval":89.4,"gpqa_diamond":58.7,"math":82.1,"swe_bench":79}},{"model":"Kimi K3","provider":"Moonshot AI","released":"2026-07","scores":{"terminal_bench_4":17.17,"swe_bench":93.4,"mmlu_pro":87.97,"gpqa_diamond":93.5,"osworld_2":58.3,"browsecomp":91.2,"frontier_code":44.2,"hle_tools":56},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, 2026-10-18 board update. The cell previously held 19.7."},"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"vendor","by":"Moonshot AI","url":"https://github.com/MoonshotAI/Kimi-K3"},"osworld_2":{"type":"vendor","by":"Moonshot AI","url":"https://github.com/MoonshotAI/Kimi-K3"},"browsecomp":{"type":"vendor","by":"Moonshot AI","url":"https://github.com/MoonshotAI/Kimi-K3"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"},"hle_tools":{"type":"vendor","by":"Moonshot AI","url":"https://github.com/MoonshotAI/Kimi-K3"}}},{"model":"Mistral Large 4","provider":"Mistral","released":"2026-10","scores":{"terminal_bench_4":27.78},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, 2026-10-18 board update. Public preview model. Mistral self-reports 28.3 on its own setup."}}},{"model":"Mistral Medium 3.5","provider":"Mistral","released":"2026-05","scores":{"swe_bench":77.6,"frontier_code":8},"provenance":{"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"}}},{"model":"Claude Opus 5","provider":"Anthropic","released":"2026-07","scores":{"terminal_bench_4":53.54,"swe_bench":97,"mmlu_pro":91.59,"gpqa_diamond":93.43,"osworld_2":70.6,"browsecomp":90.8,"frontier_code":53.4,"hle_tools":63.6},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, max effort. 1 of 198 attempts was served by the Opus 4.8 fallback; counting it as a failure gives 53.03. Vals carries this as 53.535 and its table prints 53.53."},"hle_tools":{"type":"vendor","by":"Anthropic","url":"https://www.anthropic.com/claude-opus-5-5","note":"As re-reported in the Claude Opus 5.5 launch table, the same conditions as the Opus 5.5 figure. The Opus 5 launch table printed 64.7."},"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/gpqa"},"osworld_2":{"type":"vendor","by":"Anthropic","url":"https://www.anthropic.com/claude-opus-5-system-card"},"browsecomp":{"type":"vendor","by":"Anthropic","url":"https://www.anthropic.com/claude-opus-5-system-card"},"frontier_code":{"type":"vendor","by":"Anthropic","url":"https://www.anthropic.com/claude-opus-5-system-card"}}},{"model":"Claude Fable 5","provider":"Anthropic","released":"2026-06","scores":{"terminal_bench_4":41.41,"swe_bench":95,"osworld_2":66.1,"browsecomp":87.4,"frontier_code":53.5,"hle_tools":63.9,"mmlu_pro":91.5,"gpqa_diamond":93.18},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, max effort. 42 of 198 attempts were served by the Opus 4.8 fallback; counting those as failures gives 33.84."},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/gpqa"}}},{"model":"Claude Sonnet 5","provider":"Anthropic","released":"2026-06","scores":{"terminal_bench_4":10.61,"swe_bench":85.2,"mmlu_pro":87.55,"gpqa_diamond":88.89,"browsecomp":84.7,"frontier_code":42.7,"hle_tools":57.4},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, 2026-10-18 board update. The cell previously held 9.6."},"swe_bench":{"type":"vendor","by":"Anthropic","url":"https://www.anthropic.com/claude-sonnet-5-system-card"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/gpqa"},"browsecomp":{"type":"vendor","by":"Anthropic","url":"https://www.anthropic.com/claude-sonnet-5-system-card"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"},"hle_tools":{"type":"vendor","by":"Anthropic","url":"https://www.anthropic.com/claude-sonnet-5-system-card"}}},{"model":"Claude Opus 4.8","provider":"Anthropic","released":"2026-05","scores":{"terminal_bench_4":23.23,"gpqa_diamond":93.6,"swe_bench":88.6,"osworld_2":55.7,"browsecomp":84.3,"frontier_code":46.5,"hle_tools":57.9},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, max effort."}}},{"model":"Claude Opus 4.7","provider":"Anthropic","released":"2026-04","scores":{"mmlu_pro":93.8,"human_eval":96.2,"gpqa_diamond":76.5,"math":93.1,"swe_bench":87.6,"frontier_code":38.5},"provenance":{"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"}}},{"model":"Claude Opus 4.6","provider":"Anthropic","released":"2026-03","scores":{"mmlu_pro":92.4,"human_eval":95.1,"gpqa_diamond":74.2,"math":91.8,"swe_bench":80.8,"frontier_code":26.6},"provenance":{"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"}}},{"model":"Claude Sonnet 4.6","provider":"Anthropic","released":"2026-02","scores":{"mmlu_pro":88.7,"human_eval":92,"gpqa_diamond":65.8,"math":85.4,"swe_bench":79.6,"frontier_code":24.3},"provenance":{"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"}}},{"model":"Claude Haiku 5.5","provider":"Anthropic","released":"2026-10","scores":{"terminal_bench_4":36.36},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, 2026-10-18 board update. Anthropic self-reports 39.2 at max effort on its own setup."}}},{"model":"Claude Haiku 4.5","provider":"Anthropic","released":"2026-01","scores":{"mmlu_pro":82.1,"human_eval":86.3,"gpqa_diamond":52.4,"math":74.6,"swe_bench":73.3}},{"model":"GPT-4o","provider":"OpenAI","released":"2024-05","scores":{"mmlu_pro":87.2,"human_eval":90.2,"gpqa_diamond":59.1,"math":81.3,"swe_bench":33.2}},{"model":"GPT-4.5","provider":"OpenAI","released":"2025-12","scores":{"mmlu_pro":90.1,"human_eval":93.4,"gpqa_diamond":68.7,"math":88.2,"swe_bench":38}},{"model":"o1","provider":"OpenAI","released":"2024-12","scores":{"mmlu_pro":91.8,"human_eval":94.2,"gpqa_diamond":72.5,"math":94.6,"swe_bench":48.9}},{"model":"o3-mini","provider":"OpenAI","released":"2025-01","scores":{"mmlu_pro":86.3,"human_eval":89.7,"gpqa_diamond":60.3,"math":87.1,"swe_bench":49.3}},{"model":"Gemini 2.5 Pro","provider":"Google","released":"2026-01","scores":{"mmlu_pro":91.2,"human_eval":93.8,"gpqa_diamond":71.9,"math":90.5,"swe_bench":63.8}},{"model":"Gemini 2.0 Flash","provider":"Google","released":"2025-02","scores":{"mmlu_pro":84.5,"human_eval":87.6,"gpqa_diamond":54.8,"math":77.2}},{"model":"Llama 4 Scout","provider":"Meta","released":"2025-04","scores":{"mmlu_pro":85.9,"human_eval":88.4,"gpqa_diamond":56.2,"math":79.8}},{"model":"Llama 4 Maverick","provider":"Meta","released":"2025-04","scores":{"mmlu_pro":89.3,"human_eval":91.7,"gpqa_diamond":64.1,"math":86.7,"swe_bench":24}},{"model":"Mistral Large","provider":"Mistral","released":"2025-11","scores":{"mmlu_pro":86.8,"human_eval":89.1,"gpqa_diamond":57.3,"math":80.4,"swe_bench":47.2}},{"model":"Mistral Small","provider":"Mistral","released":"2025-01","scores":{"mmlu_pro":78.4,"human_eval":82.5,"gpqa_diamond":44.6,"math":68.9}},{"model":"DeepSeek V3","provider":"DeepSeek","released":"2025-12","scores":{"mmlu_pro":88.1,"human_eval":91.2,"gpqa_diamond":63.5,"math":85.9,"swe_bench":42}},{"model":"GPT-6 Astra","provider":"OpenAI","released":"2026-09","scores":{"terminal_bench_4":59.6,"gpqa_diamond":96,"browsecomp":91.5,"frontier_code":53.3,"hle_tools":57.2},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, max effort."},"gpqa_diamond":{"type":"vendor","by":"OpenAI","url":"https://openai.com/index/gpt-6-astra/"},"browsecomp":{"type":"vendor","by":"OpenAI","url":"https://openai.com/index/gpt-6-astra/"},"frontier_code":{"type":"vendor","by":"OpenAI","url":"https://openai.com/index/gpt-6-astra/"},"hle_tools":{"type":"vendor","by":"OpenAI","url":"https://openai.com/index/gpt-6-astra/"}}},{"model":"Claude Fable 5.1","provider":"Anthropic","released":"2026-09","scores":{"terminal_bench_4":58.08,"mmlu_pro":92.38,"gpqa_diamond":93.43,"frontier_code":50.9,"hle_tools":65},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, max effort. 22 of 198 attempts were served by the Opus 5 fallback; counting those as failures gives 50.00."},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/gpqa"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"},"hle_tools":{"type":"vendor","by":"Anthropic","url":"https://www.anthropic.com/claude-fable-and-mythos-5-1"}}},{"model":"Gemini 3.8 Flash","provider":"Google","released":"2026-09","scores":{"terminal_bench_4":19.7,"swe_bench":80,"mmlu_pro":90.22,"gpqa_diamond":95.3,"frontier_code":41.2},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, 2026-10-18 board update. The cell previously held 19.19."},"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"independent","by":"Artificial Analysis","url":"https://artificialanalysis.ai/models/gemini-3-8-flash"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"}}},{"model":"Qwen3.8-Flash","provider":"Alibaba","released":"2026-08","scores":{"gpqa_diamond":91.7,"osworld_2":52.3},"provenance":{"gpqa_diamond":{"type":"vendor","by":"Alibaba","url":"https://huggingface.co/Qwen/Qwen3.8-Flash-Next"},"osworld_2":{"type":"vendor","by":"Alibaba","url":"https://huggingface.co/Qwen/Qwen3.8-Flash-Next","note":"Partial score. Binary completion on the same run is 19.4."}}},{"model":"Qwen3.7-Max","provider":"Alibaba","released":"2026-05","scores":{"swe_bench":68.8},"provenance":{"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/models/alibaba_qwen3.7-max"}}},{"model":"GLM-5.3 Flash","provider":"Z.ai","released":"2026-08","scores":{"terminal_bench_4":37.37,"swe_bench":92,"mmlu_pro":86.06,"gpqa_diamond":86.36,"frontier_code":31.8,"hle_tools":55.3},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, 2026-10-18 board update."},"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/gpqa"},"frontier_code":{"type":"independent","by":"Cognition","url":"https://cognition.com/frontiercode"},"hle_tools":{"type":"vendor","by":"Z.ai","url":"https://z.ai/blog/glm-5.3-flash"}}},{"model":"Muse Spark 1.3","provider":"Meta","released":"2026-09","scores":{"terminal_bench_4":10.61,"gpqa_diamond":94.1},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, xhigh effort. Vals lists a separate Muse Spark 1.3 Max row at 24.75."},"gpqa_diamond":{"type":"independent","by":"Artificial Analysis","url":"https://artificialanalysis.ai/evaluations/gpqa-diamond"}}},{"model":"Muse Spark 1.2","provider":"Meta","released":"2026-08","scores":{"terminal_bench_4":6.06,"swe_bench":86.6,"mmlu_pro":88.28},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, xhigh effort."},"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"}}},{"model":"Muse Spark 1.1","provider":"Meta","released":"2026-07","scores":{"swe_bench":82,"mmlu_pro":88.73,"gpqa_diamond":91.16},"provenance":{"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/gpqa"}}},{"model":"Gemini 3.5 Flash-Lite","provider":"Google","released":"2026-07","scores":{"swe_bench":75,"mmlu_pro":85.84,"gpqa_diamond":83.84},"provenance":{"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/gpqa"}}},{"model":"Gemini 3.5 Flash","provider":"Google","released":"2026-05","scores":{"terminal_bench_4":6.06,"swe_bench":78.8,"mmlu_pro":89.52,"gpqa_diamond":92.68},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, high effort."},"swe_bench":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/swebench"},"mmlu_pro":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/mmlu_pro"},"gpqa_diamond":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/gpqa"}}},{"model":"Gemini 3.1 Flash-Lite","provider":"Google","released":"2026-03","scores":{"gpqa_diamond":86.9},"provenance":{"gpqa_diamond":{"type":"vendor","by":"Google DeepMind","url":"https://deepmind.google/models/model-cards/gemini-3-1-flash-lite/"}}},{"model":"Fugu Ultra v2.0","provider":"Sakana AI","released":"2026-09","scores":{"gpqa_diamond":95.5},"provenance":{"gpqa_diamond":{"type":"vendor","by":"Sakana AI","url":"https://sakana.ai/fugu-max-release/","note":"Printed only in the benchmark chart images on the release page (fugu-ultra-v2-benchmarks.png and fugu-max-benchmarks.png); the page text has no GPQA figure. It matches the June v1 models by coincidence and is the v2 value."}}},{"model":"Fugu Max","provider":"Sakana AI","released":"2026-09","scores":{"gpqa_diamond":95.5},"provenance":{"gpqa_diamond":{"type":"vendor","by":"Sakana AI","url":"https://sakana.ai/fugu-max-release/","note":"Printed only in the benchmark chart images on the release page (fugu-ultra-v2-benchmarks.png and fugu-max-benchmarks.png); the page text has no GPQA figure. It matches the June v1 models by coincidence and is the v2 value."}}},{"model":"Hy4 preview","provider":"Tencent","released":"2026-08","scores":{"terminal_bench_4":8.08,"gpqa_diamond":92.3},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3."},"gpqa_diamond":{"type":"vendor","by":"Tencent","url":"https://huggingface.co/tencent/Hy4-preview"}}},{"model":"MAI-Code-1.1-Flash","provider":"Microsoft","released":"2026-08","scores":{"swe_bench":72.6},"provenance":{"swe_bench":{"type":"vendor","by":"Microsoft","url":"https://microsoft.ai/pdf/MAI-Code-1.1-Flash-Model-Card.PDF"}}},{"model":"MAI-Code-1-Flash","provider":"Microsoft","released":"2026-06","scores":{"swe_bench":71.6},"provenance":{"swe_bench":{"type":"vendor","by":"Microsoft","url":"https://microsoft.ai/pdf/MAI-Code-1.1-Flash-Model-Card.PDF"}}},{"model":"Gemini 4 Argon","provider":"Google","released":"2026-10","scores":{"terminal_bench_4":57.58},"provenance":{"terminal_bench_4":{"type":"independent","by":"Vals AI","url":"https://www.vals.ai/benchmarks/terminal-bench-4","note":"Vals, mini-swe-agent harness, avg@3, 2026-10-18 board update. Google self-reports 57.4 on its own setup."}}}],"provenanceNotes":{"field":"provenance","meaning":"Per-cell record of who produced a score, keyed by benchmark id.","types":{"vendor":"Published by the lab that makes the model, including figures a lab commissioned from a benchmark maintainer.","independent":"Produced by a third party that ran the model itself, such as Vals AI or Artificial Analysis."},"absent":"A benchmark id missing from a model's provenance map means provenance was not recorded, or two sources disagreed. It is not a claim that the score is unsourced.","caution":"Vendor and independent figures are not interchangeable. Observed gaps on the same model and benchmark reach 10.9 points, so do not rank across mixed provenance without checking."},"freshness":{"dataset":"benchmarks","last_updated":"2026-10-19","age_days":1,"sla_days":14,"stale":false,"reason":null,"newest_catalog_flagship":"Gemini 4 Argon","predates_flagship":false}}