Skip to content
Agents tracked: 502 Releases (7d): 613 Agent status: 4 with issuesSince last limit reset: Claude 7d · Codex 5d For AI agents: Markdown Updated Nov 18, 2026

Open models trail closed ones by 2.3 points on GPQA Diamond

Best GPQA Diamond score so far, by model release date: closed (API) and open-weight models. Live chart from AgentGid, updated daily.

Open models trail closed ones by 2.3 points on GPQA Diamond

Best GPQA Diamond score so far, by model release date: closed (API) and open-weight models

20%30%40%50%60%70%80%90%100%202420252026GPT-4, released 2023-03-14: 35.7%GPT-4 Turbo Preview, released 2023-11-06: 42.4%claude-3-opus-20240229, released 2024-02-29: 47.2%GPT-4o, released 2024-05-13: 48.9%Claude 3.5 Sonnet, released 2024-06-20: 54.0%o1-mini, released 2024-09-12: 62.4%o1, released 2024-12-17: 76.8%o3-mini, released 2025-01-31: 77.0%claude-3-7-sonnet-20250219, released 2025-02-24: 79.7%Gemini 2.5 Pro Exp, released 2025-03-25: 83.8%Gemini 2.5 Pro Preview, released 2025-06-05: 84.9%gemini-2.5-pro, released 2025-06-17: 85.3%grok-4-0709, released 2025-07-09: 87.0%GPT-5.1, released 2025-11-13: 87.6%Gemini 3 Pro Preview, released 2025-11-18: 92.6%Gemini 3.1 Pro Preview, released 2026-02-19: 94.4%GPT-5.4 Pro, released 2026-03-05: 94.6%Gemini 3.7 Flash, released 2026-08-13: 94.8%Gemini 3.8 Flash, released 2026-09-02: 95.4%GPT-6 Astra, released 2026-09-03: 95.8%Gemini 4 Argon, released 2026-10-13: 96.4%Gemini 4 Carbon, released 2026-11-10: 96.5%Closed 96.5%Gemini 4 Carbonllama-2-70b-chat-hf, released 2023-07-18: 26.3%mixtral-8x7b-instruct-v0.1, released 2023-12-11: 30.6%dbrx-instruct, released 2024-03-27: 32.9%wizardlm-2-8x22b, released 2024-04-15: 43.4%llama-3.1-70b-instruct, released 2024-07-23: 44.2%llama-3.1-405b-instruct, released 2024-07-23: 50.9%mistral-large-2411, released 2024-11-18: 51.3%phi-4, released 2024-12-12: 56.1%deepseek-v3, released 2024-12-26: 56.5%deepseek-r1-distill-qwen-32b, released 2025-01-20: 64.1%deepseek-r1, released 2025-01-20: 71.7%DeepSeek-R1, released 2025-05-28: 76.3%qwen3-235b-a22b-thinking-2507, released 2025-07-25: 80.0%Kimi K2 Thinking Turbo, released 2025-11-06: 84.2%Kimi K2.5, released 2026-01-27: 87.6%GLM-5, released 2026-02-11: 87.8%GLM-5.1, released 2026-04-07: 89.9%Kimi K2.6, released 2026-04-20: 90.8%DeepSeek v4 Pro, released 2026-04-24: 90.9%GLM-5.2, released 2026-06-16: 91.9%Kimi K3, released 2026-07-16: 93.1%DeepSeek V4.1 Flash 1017, released 2026-10-17: 93.2%Kimi K3.1 Pro, released 2026-10-23: 94.2%Open-weight 94.2%Kimi K3.1 Pro
Each step is a model that beat the best score of its kind so far. Openness as Epoch AI records it. Source: Epoch AI benchmarking hub (CC BY 4.0) · Nov 17, 2026.

Put this chart on your site

The chart updates by itself. Please keep the link to the source.

More charts