Benchmark leaderboard - 25 entries
SimpleQA leaderboard
SimpleQA is a benchmark that evaluates the ability of language models to answer short, fact-seeking questions. As of October 17, 2026, gpt-4.5-preview-2025-02-27 ranks first on the SimpleQA board with 62.5.
Last verified from github.com. Read the caveats
SimpleQA
25 entries, in the board's order
| 1 | gpt-4.5-preview-2025-02-27 simple-evals, assistant system prompt ("You are a helpful assistant."), zero-shot Official leaderboard | 62.5 |
| 2 | o3 simple-evals, default (medium) reasoning, no system prompt, no tools enabled Official leaderboard | 49.4 |
| 3 | o3-low simple-evals, low reasoning effort, no system prompt, no tools enabled Official leaderboard | 49.4 |
| 4 | o3-high simple-evals, high reasoning effort, no system prompt, no tools enabled Official leaderboard | 48.6 |
| 5 | o1 simple-evals, no system prompt (n/a for o-series) Official leaderboard | 42.6 |
| 6 | o1-preview simple-evals, no system prompt (n/a for o-series) Official leaderboard | 42.4 |
| 7 | gpt-4.1-2025-04-14 simple-evals, assistant system prompt, zero-shot Official leaderboard | 41.6 |
| 8 | gpt-4o-2024-08-06 simple-evals, assistant system prompt, zero-shot Official leaderboard | 40.1 |
| 9 | gpt-4o-2024-05-13 simple-evals, assistant system prompt, zero-shot Official leaderboard | 39.0 |
| 10 | gpt-4o-2024-11-20 simple-evals, assistant system prompt, zero-shot Official leaderboard | 38.8 |
| 11 | Kimi K2 Instruct Metric: Correct; instruct model evaluation table (non-thinking) Vendor-reported | 31.0 |
| 12 | DeepSeek-R1 SimpleQA (Correct); max generation length 32,768 tokens, temperature 0.6, top-p 0.95 Vendor-reported | 30.1 |
| 13 | Claude 3.5 Sonnet simple-evals "Other Models (Reported)" row, prompt: unknown Official leaderboard | 28.9 |
| 14 | DeepSeek-V3 SimpleQA (Correct); chat model evaluation table Vendor-reported | 24.9 |
| 15 | gpt-4-turbo-2024-04-09 simple-evals, assistant system prompt, zero-shot Official leaderboard | 24.2 |
| 16 | Claude 3 Opus simple-evals "Other Models (Reported)" row, prompt: unknown Official leaderboard | 23.5 |
| 17 | o4-mini simple-evals, default (medium) reasoning, no system prompt, no tools enabled Official leaderboard | 20.2 |
| 18 | o4-mini-low simple-evals, low reasoning effort, no system prompt, no tools enabled Official leaderboard | 20.2 |
| 19 | o4-mini-high simple-evals, high reasoning effort, no system prompt, no tools enabled Official leaderboard | 19.3 |
| 20 | MiniMax-M1-80K Extended thinking 80K Vendor-reported | 18.5 |
| 21 | MiniMax-M1-40K Extended thinking 40K Vendor-reported | 17.9 |
| 22 | gpt-4.1-mini-2025-04-14 simple-evals, assistant system prompt, zero-shot Official leaderboard | 16.8 |
| 23 | o3-mini-high simple-evals, high reasoning effort, no system prompt Official leaderboard | 13.8 |
| 24 | o3-mini simple-evals, default (medium) reasoning, no system prompt Official leaderboard | 13.4 |
| 25 | o3-mini-low simple-evals, low reasoning effort, no system prompt Official leaderboard | 13.0 |
About this board
- Rows marked vendor-reported come from the model maker's own model card or launch post, run under its own settings. Read the setting under each model before comparing two rows.
- No live official SimpleQA leaderboard exists. The results table in openai/simple-evals (README) is OpenAI's only official one, and it has been frozen since a July 2025 deprecation notice. Its rows carry date '2025-07' (when the table was frozen), not the date of each individual result.
- The rows marked Official leaderboard come from OpenAI's simple-evals table. The Claude 3.5 Sonnet and Claude 3 Opus rows sit in that table's 'Other Models (Reported)' section, so OpenAI is only relaying numbers it did not measure itself.
- Most current frontier vendors no longer report original SimpleQA. Google's Gemini 3.1 Pro page (deepmind.google/models/gemini/pro/, read 2026-10-17) lists no SimpleQA row; Google now reports SimpleQA Verified, which is a separate benchmark and was not included.
- Gemini 2.5 Pro: the 2025-03-25 Google blog post shows its benchmark table only as an image, and the model card PDF returned 403, so no first-party number could be confirmed.
- Gemini 2.5 Pro (06-05) 54.0 and OpenAI-o3 49.4 appear in MiniMax's MiniMax-M1 README. These are third-party figures inside another vendor's table, so they were left out.
- Dropped from the top 25 because they scored lower (all simple-evals official): gpt-4o-mini-2024-07-18 09.5, o1-mini 07.6, gpt-4.1-nano-2025-04-14 07.6.
- Kimi K2 Base 35.3 (5-shot, base model) is not included because it is a pretrained base model, not comparable with the zero-shot chat models.
FAQ
What does SimpleQA measure?
SimpleQA is a benchmark that evaluates the ability of language models to answer short, fact-seeking questions.
Which model leads SimpleQA?
As of October 17, 2026, gpt-4.5-preview-2025-02-27 ranks first on the SimpleQA board with 62.5. That is the board's own order, not a Developers Digest ranking.
Where do these SimpleQA scores come from?
20 of the 25 entries come from the official leaderboard at github.com, and 5 are vendor-reported, taken from the model maker's own model card or launch post. Every entry links to its source, which was read on October 17, 2026, and lists the setting it was run under.
Related reading
Sources
Last verified . Each row links to the page its score was copied from.
Newsletter
Get new benchmark results as they land
One email per week. New model releases, the leaderboard moves that matter and what they mean for what you build.