benchgap
LLM benchmark leaderboard · gaps filled

LLM benchmark scores, with the gaps filled

benchgap is an LLM benchmark leaderboard that fills in the missing scores. Most models are only ever run on a handful of benchmarks, so benchgap calibrates benchmarks against each other on the models measured on both, then estimates each missing score with its cross-validated error and a confidence level. Measured and estimated scores are always marked apart.

As of 2026-11-21: 612 models, 240 benchmark versions, 8226 measured and 27521 estimated scores (5081 high, 9813 medium and 12627 low confidence). Measured scores come from public leaderboards (artificial-analysis, published, vals-ai).

AA-HLE leaderboard

As of 2026-11-21, the highest measured score on AA-HLE is 74.3% by Claude Fable 5.2. 97 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Claude Fable 5.274.3%measured
2Claude Opus 5.561.4%measured
3GPT-6.2 Astra61.4%estimated ± 3.9 pp, medium confidence
4Claude Fable 5.159.1%measured
5Gemini 4 Neon59.0%estimated ± 9.4 pp, medium confidence
6Claude Mythos 557.9%estimated ± 4.6 pp, high confidence
7Gemini 4 Argon57.1%measured
8GPT-6.2 Sol56.1%estimated ± 3.9 pp, high confidence
9Claude Fable 555.5%measured
10Claude Sonnet 5.555.0%measured
11Claude Opus 554.9%measured
12GPT-6 Astra54.7%measured
13Gemini 4 Hydrogen53.7%estimated ± 9.4 pp, medium confidence
14Sakana Fugu53.7%estimated ± 3.9 pp, high confidence
15Sakana Fugu-Ultra53.7%estimated ± 3.9 pp, high confidence
16Muse Wonder Full53.6%measured
17GPT-6.1 Sol52.9%measured
18Gemini 4 Carbon52.8%estimated ± 6.0 pp, medium confidence
19Kimi K3.1 Pro52.5%measured
20Qwen4 Max50.5%estimated ± 4.6 pp, high confidence
21Muse Wonder50.1%measured
22GPT-5.6 Sol49.5%measured
23MiMo-V2.6-Pro49.4%measured
24Claude Opus 4.848.7%measured
25Muse Spark 1.348.7%measured
26MiniMax M3.1 Flash48.0%measured
27Gemini 3.7 Flash47.9%measured
28GPT-6 Sol47.9%measured
29Gemini 3.8 Flash47.8%measured
30Pareto 26.947.8%estimated ± 4.6 pp, high confidence
31dots3-note Preview47.7%estimated ± 6.0 pp, medium confidence
32Hy447.1%estimated ± 6.5 pp, medium confidence
33Gemini 3.1 Pro47.0%measured
34Kimi K346.9%measured
35Step 5 Preview46.5%measured
36GPT-6.1 Terra46.2%measured
37Muse Spark 1.146.2%measured
38GPT-5.545.8%measured
39Sage-A1.545.7%estimated ± 5.8 pp, medium confidence
40Muse Spark 1.245.5%measured
41ERNIE 5.5 Flash Preview45.2%estimated ± 4.6 pp, high confidence
42Hy4 flash44.8%estimated ± 6.5 pp, medium confidence
43Claude Haiku 5.544.4%measured
44Inkling 2 Small44.1%measured
45Muse Spark Open43.9%measured
46GPT-5.443.7%measured
47Grok 4.743.1%measured
48Qwen3.8 Max Preview43.1%measured
49Ornith-1.5-397B43.1%estimated ± 3.9 pp, high confidence
50GPT-5.6 Terra42.9%measured
51Grok 4.642.9%measured
52Gemini 3.5 Flash42.7%measured
53Grok 4.542.7%measured
54GPT-5.3 Codex42.5%measured
55Qwen3.8 Max42.4%estimated ± 3.9 pp, high confidence
56Claude Opus 4.742.3%measured
57GLM-5.342.3%measured
58GPT-5.5 Pro42.0%estimated ± 4.6 pp, high confidence
59Pareto 26.10 Preview41.8%estimated ± 3.9 pp, high confidence
60GPT-5.4 Pro41.7%estimated ± 4.6 pp, high confidence
61Hy4 preview41.5%estimated ± 3.9 pp, high confidence
62Claude Sonnet 541.3%measured
63GLM-5.241.1%measured
64DeepSeek V4 Pro41.0%measured
65Gemini 3.6 Flash40.8%measured
66Muse Spark40.7%measured
67Qwen3.7 Max40.5%measured
68Agents-A140.2%estimated ± 6.5 pp, medium confidence
69Claude Opus 4.6 (Adaptive)39.9%measured
70Gemini 3 Pro39.7%measured
71GPT-5.6 Luna39.5%measured
72Ling 3.1 Flash39.4%measured
73Muse Ember39.4%estimated ± 9.4 pp, medium confidence
74DeepSeek V4.1 Flash 101739.2%measured
75MiniMax M339.0%measured
76DeepSeek V4 Flash 073138.6%measured
77GPT-6 Luna38.5%measured
78Qwen3.8-Flash-Next38.0%measured
79GPT-5.237.7%measured
80Qwen3.8-Omni-Flash37.6%estimated ± 3.9 pp, high confidence
81Kimi K2.637.5%measured
82Grok 4.337.2%measured
83Mistral Large 436.9%estimated ± 5.8 pp, medium confidence
84Particle-136.0%estimated ± 3.9 pp, high confidence
85Beam 1.0 501B-A23B FP835.7%estimated ± 3.9 pp, high confidence
86GPT-5.2-Codex35.7%measured
87MiMo-V2.5-Pro35.7%measured
88Qwen3.7 Plus35.6%measured
89MiMo-V2.6-Flash35.1%measured
90Kimi K2.7 Code35.0%measured
91Interfaze Beta34.7%estimated ± 3.9 pp, high confidence
92Beam 1.0 501B-A23B34.7%measured
93Apodex 1.134.1%measured
94Apodex 1.1 Mini34.1%measured
95Qwen3.8-27B33.9%measured
96Hy333.5%measured
97Hy3 Preview33.5%measured
98Inkling-Small33.3%measured
99Beam 1.0 501B-A23B NVFP433.3%estimated ± 3.9 pp, high confidence
100Mercury 333.1%estimated ± 3.9 pp, high confidence
101Ornith-1.5-35B-A3B33.1%estimated ± 3.9 pp, high confidence
102Nemotron 3.5 Bolt 74B-A9B BF1632.2%estimated ± 3.9 pp, high confidence
103Inkling31.9%measured
104Grok 4.2031.5%estimated ± 3.9 pp, high confidence
105OrcaCyber-Nano-1.031.0%estimated ± 3.9 pp, high confidence
106Qwen 3.6 Max (preview)30.8%measured
107Kimi K2.530.7%measured
108Nemotron 3.5 Super VL 120B A12B30.7%measured
109Nemotron 3.5 Super VL 120B A12B NVFP430.7%measured
110MiMo-V2-Pro30.4%measured
111Claude Opus 4.5 Thinking30.1%measured
112GLM-5.130.1%measured
113GLM-5.3-Flash29.9%estimated ± 6.2 pp, medium confidence
114Solar Pro 429.8%measured
115Claude Sonnet 4.5 Thinking29.6%estimated ± 6.0 pp, medium confidence
116A.X K229.6%measured
117MiniMax M2.729.6%measured
118GLM-529.3%measured
119Qwen3.5 397B29.0%measured
120GPT-528.5%measured
121GPT-5.128.5%measured
122Nemotron 3 Ultra28.4%measured
123GPT-5.4 nano28.3%measured
124GPT-5.4 mini28.1%measured
125GLM-5-Turbo27.8%measured
126Qwen3.6 Plus27.8%measured
127Inkling 2 Mini27.7%estimated ± 3.9 pp, high confidence
128GLM-4.727.4%measured
129Ornith-1.5-9B27.3%estimated ± 3.9 pp, high confidence
130Solar Open 227.1%estimated ± 3.9 pp, high confidence
131Grok 426.7%measured
132Seed 2.1 Pro26.6%estimated ± 7.6 pp, low confidence
133Ternary Bonsai 2 27B26.2%estimated ± 3.9 pp, high confidence
134Seed 2.1 Turbo26.1%estimated ± 7.6 pp, low confidence
135GPT-5.1-Codex25.7%measured
136GPT-5.1-Codex-Max25.7%measured
137Ripple-65B Preview25.2%estimated ± 3.9 pp, high confidence
138Qwen3.5-122B-A10B25.2%measured
139Nemotron 3.5 Bolt 74B-A9B24.7%measured
140o3-pro24.0%estimated ± 2.7 pp, high confidence
141Qwen3.5-27B23.9%measured
142Ling 3.0 Flash23.7%measured
143MAI-Thinking-123.6%estimated ± 3.9 pp, high confidence
144Gemma 4 31B23.6%measured
145Ling 3.0 Flash FP823.3%estimated ± 3.9 pp, high confidence
146Qwen3.6-27B23.1%measured
147Gemini 2.5 Pro22.5%measured
148Claude 4.1 Opus22.4%estimated ± 9.4 pp, medium confidence
149Qwen3.5 Flash22.4%estimated ± 6.2 pp, medium confidence
150Qwen3.6-35B-A3B22.2%measured
151MiMo-V2-Omni22.1%measured
152Ling 3.0 Flash VL22.0%measured
153Muse Glimmer 30B22.0%measured
154Step 3.7 Flash21.4%measured
155Qwen3.5-35B-A3B21.0%measured
156MiMo-V2.520.2%estimated ± 6.2 pp, medium confidence
157o320.1%measured
158Claude Sonnet 4.519.9%estimated ± 5.7 pp, medium confidence
159GPT-OSS 120B19.6%measured
160Gemini 3.1 Flash-Lite19.4%estimated ± 6.2 pp, medium confidence
161Gemma 4 26B A4B19.3%measured
162Grok 4.1 Fast (Reasoning)19.3%measured
163Claude Opus 4.619.1%measured
164Grok 4 Fast (Reasoning)19.1%measured
165Gemini 3.5 Flash-Lite18.8%measured
166Quasar 438B18.7%measured
167K-EXAONE 2.018.6%measured
168GLM-5V-Turbo17.1%measured
169DeepSeek-R1-052817.0%estimated ± 5.7 pp, medium confidence
170o4-mini16.5%measured
171Grok 3 Mini16.3%estimated ± 5.7 pp, medium confidence
172DeepSeek-R115.8%measured
173Trinity-Large-Preview15.8%measured
174Trinity-Large-Thinking15.8%measured
175Gemma 4 12B15.7%measured
176Nemotron 3 Super 100B15.2%estimated ± 5.7 pp, medium confidence
177Spark-X2.6-4B15.0%estimated ± 3.9 pp, high confidence
178o1-pro15.0%estimated ± 5.7 pp, medium confidence
179Gemini 3 Flash15.0%measured
180K-Exaone13.9%measured
181Mistral Medium 3.5 128B13.8%measured
182Qwen3 235B 250713.7%estimated ± 5.7 pp, medium confidence
183Claude Sonnet 4.613.3%measured
184Claude Opus 4.513.2%measured
185LFM2.5-24B-A2B12.6%estimated ± 3.9 pp, high confidence
186Claude 4 Opus Thinking12.3%measured
187o1-preview12.1%estimated ± 2.7 pp, high confidence
188Grok 3 [Beta]12.0%estimated ± 5.7 pp, medium confidence
189Command A+12.0%measured
190Qwen3 Max11.9%measured
191GLM-4.7-Flash11.9%estimated ± 5.7 pp, medium confidence
192Mercury 2.511.8%measured
193DeepSeek V3.211.2%measured
194Granite 4.2 30B11.2%measured
195North Mini Code11.1%measured
196GPT-OSS 20B11.0%measured
197Sarvam 105B11.0%measured
198Nemotron 3.5 Lightning 30B A3B NVFP410.6%measured
199Solar Pro 310.3%measured
200ZAYA1-8B10.1%estimated ± 3.9 pp, high confidence
201Mistral Small 49.9%measured
202Qwen3.5-4B9.9%measured
203Magistral Medium9.8%measured
204Claude 3.7 Sonnet Thinking9.7%measured
205Granite 4.2 8B9.7%measured
206Claude Haiku 4.59.3%estimated ± 6.2 pp, medium confidence
207GLM-4.59.3%estimated ± 6.2 pp, medium confidence
208Qwen3 235B-A22B (Reasoning)9.3%estimated ± 5.7 pp, medium confidence
209Ling 3.0 Tiny9.3%measured
210LongCat-Flash-Lite-Sparse9.1%estimated ± 3.9 pp, high confidence
211MiniCPM5-2B8.9%measured
212MiMo-V2-Flash8.6%measured
213GPT-4.58.5%estimated ± 5.7 pp, medium confidence
214Phi-4-reasoning-plus8.2%estimated ± 5.7 pp, medium confidence
215Grok Code Fast 18.0%measured
216Qwen3 32B (Reasoning)7.9%estimated ± 5.7 pp, medium confidence
217o3-mini7.9%measured
218Magistral Small7.6%measured
219Qwen3-Omni-30B-A3B-Thinking7.5%measured
220Sarvam 30B7.5%measured
221Kimi K27.4%measured
222Nemotron Ultra 253B7.4%measured
223QwQ-32B7.4%measured
224GLM-4.5-Air7.0%measured
225o17.0%measured
226LFM2.5-8B-A1B6.9%measured
227Celeris-16.8%measured
228DeepSeek V3.16.7%measured
229Granite 4.2 3B6.6%measured
230DeepSeek R1 Distill Llama 70B6.5%estimated ± 5.7 pp, medium confidence
231Granite-4.0-H-350M6.4%measured
232Ling 2.6 Flash6.3%measured
233Mellum2.1-12B-A2.5B-Thinking6.3%estimated ± 3.9 pp, high confidence
234Claude 4 Opus6.2%measured
235LFM2.5-2.6B6.2%measured
236MiniCPM5-1B6.1%estimated ± 6.7 pp, medium confidence
237Exaone 4.0 1.2B5.7%measured
238LFM3-1.7B5.7%estimated ± 3.9 pp, high confidence
239Qwen3 235B-A22B5.6%estimated ± 5.7 pp, medium confidence
240GLM-4.65.5%measured
241Granite-4.0-350M5.5%measured
242Grok 4.1 Fast5.1%measured
243LFM2.5-VL-1.6B-Extract5.1%measured
244Exaone 4.0 32B5.0%measured
245GPT-4.1 mini5.0%measured
246Granite-4.0-H-1B5.0%measured
247Phi-4 Multimodal Instruct5.0%measured
248DeepSeek-R1-0528-Qwen3-8B5.0%estimated ± 5.7 pp, medium confidence
249Llama 4 Maverick4.9%measured
250Gemma 4 E2B4.8%measured
251Nemotron 3 Nano Omni 30B A3B4.8%measured
252Laguna XS.24.8%estimated ± 6.2 pp, medium confidence
253DeepSeek V3 03244.7%measured
254Gemini 2.5 Flash4.7%measured
255Qwen2.5-Max4.7%estimated ± 5.7 pp, medium confidence
256DeepSeek R1 Distill Qwen 32B4.6%measured
257Gemini 1.5 Pro4.6%measured
258Nemotron 3 Nano 30B4.6%measured
259Qwen3-Omni-30B-A3B-Instruct4.6%measured
260Claude 3.5 Sonnet4.4%estimated ± 5.7 pp, medium confidence
261Gemma 3 27B4.4%measured
262DeepSeek R1 Distill Qwen 14B4.3%estimated ± 5.7 pp, medium confidence
263DeepSeek-V2.54.3%estimated ± 2.7 pp, high confidence
264Claude 4 Sonnet4.3%measured
265Gemini 2.0 Flash4.3%measured
266Llama 3.1 8B4.3%measured
267Mistral Small 3.14.3%measured
268Laguna M.14.3%estimated ± 6.2 pp, low confidence
269Claude 3.7 Sonnet4.2%measured
270Gemini 1.0 Pro4.2%measured
271GPT-4.14.2%measured
272GPT-4o mini4.2%measured
273Mistral Large 34.2%measured
274Claude 3 Haiku4.1%measured
275Mistral Medium 34.1%measured
276Llama 3.1 405B4.0%measured
277Devstral Small3.8%measured
278Gemma 4 E4B3.8%measured
279GPT-4.1 nano3.8%measured
280Llama 4 Scout3.8%measured
281Phi-43.8%measured
282Phi-4-reasoning3.8%measured
283Gemini 2.5 Flash-Lite3.7%measured
284Solar Pro 23.7%measured
285Claude 3.5 Haiku3.6%measured
286Llama 3.1 70B3.6%measured
287Llama 3.3 70B3.6%measured
288o1-mini3.6%measured
289Ultravox v0.6 Llama 3.3 70B3.6%measured
290Qwen2.5 Coder 32B Instruct3.5%measured
291Grok 23.4%estimated ± 5.7 pp, medium confidence
292Mistral Large 23.3%measured
293Gemini 1.5 Flash3.2%measured
294Nova Pro3.2%measured
295Mellum2-12B-A2.5B-Thinking3.2%estimated ± 3.9 pp, medium confidence
296GPT-4 Turbo3.1%measured
297Qwen3 32B3.1%estimated ± 5.7 pp, medium confidence
298ZAYA1-74B-Preview3.1%estimated ± 3.9 pp, medium confidence
299MiniMax-Text-013.0%estimated ± 5.7 pp, medium confidence
300DeepSeek V32.9%measured
301Claude 3 Opus2.8%measured
302GPT-4o2.4%measured
303Grok 2 mini2.2%estimated ± 5.7 pp, medium confidence
304Gemini 2.0 Flash-Lite2.1%estimated ± 5.7 pp, medium confidence
305Qwen2.5-VL-32B1.2%estimated ± 5.7 pp, medium confidence
306Granite-4.0-1B0.0%estimated ± 5.7 pp, medium confidence
307LFM2.5-230M0.0%estimated ± 5.7 pp, medium confidence
308LFM2.5-350M0.0%estimated ± 5.7 pp, medium confidence
309LFM2.5-VL-450M0.0%estimated ± 5.7 pp, medium confidence
310LLaDA2.2-mini0.0%estimated ± 3.9 pp, medium confidence
311Mellum2-12B-A2.5B-Instruct0.0%estimated ± 3.9 pp, medium confidence
312Phi-3.5-mini0.0%estimated ± 5.7 pp, medium confidence
313Soofi S 30B-A3B0.0%estimated ± 3.9 pp, medium confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General