Knowledge

SimpleQA Leaderboard

SimpleQA measures factual accuracy on short, fact-seeking questions with a single correct answer — directly probing how often a model is right versus confidently wrong (hallucination) on simple facts.

Source: epoch15 open models ranked+63 proprietaryData through Sep 2026

All models ranked on SimpleQA

Proprietary / closed models are shown dimmed — you can't run them locally, but they show where the open field stands.

#ModelScore
1GPT 6 Astra Max · proprietary
75.6%
2Gemini 3.1 Pro Preview (high) · proprietary
73.5%
3Claude Fable 5.1 Max · proprietary
70.8%
4Claude Fable 5 (xhigh) · proprietary
70.7%
5GPT 5.6 Sol Max · proprietary
69.7%
6Gemini 3.7 Flash (high) · proprietary
69.2%
7Gemini 3 Flash Preview (high) · proprietary
66.8%
8Gemini 3.5 Flash (high) · proprietary
66.2%
9Gemini 3.6 Flash (high) · proprietary
66.2%
10GPT 5.5 (xhigh) · proprietary
63.0%
11Muse Spark 1.2 (xhigh) · proprietary
60.3%
12Claude Opus 5 Max · proprietary
59.9%
13Muse Spark 1.1 · proprietary
57.8%
14Qwen3.7 Max · proprietary
55.8%
15Claude Opus 4.8 Max · proprietary
53.0%
16DeepSeek V4 Pro 0813 · 1650.5B
52.9%
17Qwen3.6 Max Preview · proprietary
52.0%
18Claude Opus 4.7 (xhigh) · proprietary
51.7%
19Kimi K3 · 2779.9B
50.6%
20GPT 5 (Aug 07, 2025, high) · proprietary
50.1%
21O3 (Apr 16, 2025, high) · proprietary
49.4%
22Grok 4.6 (high) · proprietary
49.3%
23Grok 4.6 (xhigh) · proprietary
48.9%
24Qwen3 Max (Sep 23, 2025) · proprietary
48.8%
25Grok 4.5 (high) · proprietary
48.3%
26GPT 5.1 (Nov 13, 2025, high) · proprietary
48.0%
27Claude Opus 4.6 Max · proprietary
47.0%
28DeepSeek V4 Pro · 1598.8B
47.0%
29GPT 5.4 Pro (Mar 05, 2026, xhigh) · proprietary
46.3%
30Qwen3.8 Max (xhigh) · proprietary
45.8%
31Claude Opus 4.5 (Nov 01, 2025, 32K) · proprietary
45.7%
32GPT 5.4 (Mar 05, 2026, xhigh) · proprietary
45.1%
33Qwen3.6 Plus · proprietary
44.1%
34GPT 5.6 Terra Max · proprietary
43.2%
35O1 (Dec 17, 2024, high) · proprietary
41.1%
36GLM 5.3 · 753.3B
41.0%
37GPT 5.6 Luna Max · proprietary
41.0%
38Qwen3 235B A22B Thinking 2507 · 235.1B
40.4%
39Inkling · 952.4B
40.3%
40GPT 5.2 (Dec 11, 2025, xhigh) · proprietary
37.1%
41Kimi K2.7 Code · 1026.9B
36.5%
42Claude Sonnet 4.6 (high) · proprietary
35.5%
43Kimi K2.6 · 1026.9B
34.9%
44GPT 5.2 (Dec 11, 2025, high) · proprietary
34.3%
45Kimi K2.5 · 1026.9B
34.3%
46GLM 5.2 · 753.3B
34.2%
47GLM 5.1 · 753.9B
34.0%
48Claude Sonnet 5 Max · proprietary
33.7%
49DeepSeek V4 Flash 0731 · 304.2B
33.6%
50Grok 4.3 (high) · proprietary
33.2%
51Claude Sonnet 5 (xhigh) · proprietary
32.9%
52Claude Sonnet 4.6 Max · proprietary
32.8%
53GPT 5.2 (Dec 11, 2025, low) · proprietary
32.8%
54GPT 5.2 (Dec 11, 2025, medium) · proprietary
32.7%
55GLM 4.7 · 358.3B
32.2%
56GPT 4.1 (Apr 14, 2025) · proprietary
31.1%
57Claude Sonnet 4.5 (Sep 29, 2025, 59K) · proprietary
30.7%
58Grok 4.20 0309 Reasoning · proprietary
30.2%
59GPT 5.4 Mini (Mar 17, 2026, high) · proprietary
29.4%
60GPT 4o (Aug 06, 2024) · proprietary
26.0%
61Qwen3.5 Plus · proprietary
25.4%
62Claude Sonnet 4.5 (Sep 29, 2025) · proprietary
23.7%
63GPT 5 Mini (Aug 07, 2025, high) · proprietary
21.6%
64Qwen3.5 Flash · proprietary
20.3%
65O4 Mini (Apr 16, 2025, low) · proprietary
19.6%
66Inkling Small · 266.0B
19.1%
67O4 Mini (Apr 16, 2025, high) · proprietary
18.8%
68Qwen3.6 Flash · proprietary
15.9%
69O3 Mini (Jan 31, 2025, high) · proprietary
15.3%
70Claude Haiku 4.5 (Oct 01, 2025) · proprietary
13.2%
71GPT 4.1 Mini (Apr 14, 2025) · proprietary
12.7%
72Claude 3 Opus (Feb 29, 2024) · proprietary
12.6%
73Claude Haiku 4.5 (Oct 01, 2025, 32K) · proprietary
12.6%
74GPT 5 Nano (Aug 07, 2025, high) · proprietary
11.7%
75GPT 5.4 Nano (Mar 17, 2026, high) · proprietary
11.7%
76Gemma 4 31B IT · 31.3B
10.4%
77GPT 4o Mini (Jul 18, 2024) · proprietary
8.3%
78GPT 4.1 Nano (Apr 14, 2025) · proprietary
6.0%

Score vs model size

Which models give the most quality for their size — the ones worth running locally.

100B1Tmodel size (log scale) →52.9%10.4%Kimi K3 · 2.8T · 50.6%Inkling · 952B · 40.3%Kimi K2.7 Code · 1T · 36.5%Kimi K2.6 · 1T · 34.9%Kimi K2.5 · 1T · 34.3%GLM 5.2 · 753B · 34.2%GLM 5.1 · 754B · 34.0%DeepSeek V4 Flash 0731 · 304B · 33.6%GLM 4.7 · 358B · 32.2%Inkling Small · 266B · 19.1%Gemma 4 31B IT · 31B · 10.4%Gemma 4 31B ITQwen3 235B A22B Thinking 2507 · 235B · 40.4%Qwen3 235B A22B Think…GLM 5.3 · 753B · 41.0%GLM 5.3DeepSeek V4 Pro · 1.6T · 47.0%DeepSeek V4 ProDeepSeek V4 Pro 0813 · 1.7T · 52.9%DeepSeek V4 Pro 0813
Each dot is a model. Up = higher score, left = smaller (easier to run locally). The dashed line marks the efficiency frontier — the best score you can get at each size or smaller.
  • Gemma 4 31B IT, 31B, score 10.4% — on the efficiency frontier (best score at its size or smaller).
  • Qwen3 235B A22B Thinking 2507, 235B, score 40.4% — on the efficiency frontier (best score at its size or smaller).
  • GLM 5.3, 753B, score 41.0% — on the efficiency frontier (best score at its size or smaller).
  • DeepSeek V4 Pro, 1.6T, score 47.0% — on the efficiency frontier (best score at its size or smaller).
  • DeepSeek V4 Pro 0813, 1.7T, score 52.9% — on the efficiency frontier (best score at its size or smaller).

SimpleQA: frequently asked questions

What is the best open LLM on SimpleQA?
DeepSeek V4 Pro 0813 is the top open model on SimpleQA, scoring 52.9%. Among all models tested — including proprietary ones — it ranks #16. The top model overall is GPT 6 Astra Max (OpenAI) at 75.6%.
What's the best SimpleQA model you can run on a 24 GB GPU?
Gemma 4 31B IT is the highest-scoring open model that fits in 24 GB at 4-bit quantization (about 17 GB), scoring 10.4% on SimpleQA.
Can open models match proprietary models on SimpleQA?
Not quite on SimpleQA: the strongest proprietary model (GPT 6 Astra Max) scores 75.6%, ahead of the best open model (DeepSeek V4 Pro 0813) at 52.9% — but you can run the open one yourself.

Scores aggregated from epoch. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.