Coding
ALE-Bench Leaderboard
ALE-Bench, from Sakana AI and AtCoder, tests long-horizon algorithmic optimisation using problems drawn from AtCoder Heuristic Contests, where there is no single correct answer and solutions are scored on quality. Results are reported as an AtCoder-style performance rating, where higher is better.
Source: epoch51 open models ranked+87 proprietaryData through Nov 2026
All models ranked on ALE-Bench
Proprietary / closed models are shown dimmed — you can't run them locally, but they show where the open field stands.
| # | Model | Score | VRAM @ Q4 |
|---|---|---|---|
| 1 | GPT 6 Astra Max · proprietary | 2951.3 | API only |
| 2 | GPT 6.1 Sol Max · proprietary | 2743.8 | API only |
| 3 | GPT 6.1 Terra Max · proprietary | 2608.8 | API only |
| 4 | Claude Fable 5.2 (high) · proprietary | 2509.8 | API only |
| 5 | GPT 6 Sol Max · proprietary | 2462.0 | API only |
| 6 | GPT 5.6 Sol Max · proprietary | 2176.9 | API only |
| 7 | Claude Opus 5 (high) · proprietary | 2164.6 | API only |
| 8 | Claude Opus 5.5 (high) · proprietary | 2147.4 | API only |
| 9 | Claude Fable 5.1 (high) · proprietary | 2143.2 | API only |
| 10 | Gemini 4 Argon (high) · proprietary | 2100.1 | API only |
| 11 | Claude Fable 5 (high) · proprietary | 2041.3 | API only |
| 12 | GPT 5.6 Terra Max · proprietary | 1951.4 | API only |
| 13 | GPT 5.5 (xhigh) · proprietary | 1943.0 | API only |
| 14 | Kimi K3.1 Pro · 2780.1B | 1932.8 | ~1674 GB |
| 15 | Claude Sonnet 5.5 (high) · proprietary | 1819.1 | API only |
| 16 | Gemini 4 Argon (low) · proprietary | 1705.4 | API only |
| 17 | Muse Spark 1.3 (xhigh) · proprietary | 1688.3 | API only |
| 18 | Qwen3.8 Max (Sep 02, xhigh) · proprietary | 1679.1 | API only |
| 19 | GPT 5.6 Luna Max · proprietary | 1667.4 | API only |
| 20 | GPT 5.3 Codex (xhigh) · proprietary | 1655.2 | API only |
| 21 | DeepSeek V4.1 Flash 1017 · 763.2B | 1622.0 | ~458 GB |
| 22 | GPT 5.4 (Mar 05, 2026, high) · proprietary | 1607.0 | API only |
| 23 | Muse Spark Open 816B A42B · 822.0B | 1590.1 | ~543 GB |
| 24 | GPT 5.5 (medium) · proprietary | 1589.4 | API only |
| 25 | Claude Haiku 5.5 (high) · proprietary | 1587.0 | API only |
| 26 | GPT 6 Luna (xhigh) · proprietary | 1576.9 | API only |
| 27 | Claude Opus 4.8 (high) · proprietary | 1563.8 | API only |
| 28 | Grok 4.7 (xhigh) · proprietary | 1545.8 | API only |
| 29 | Kimi K3 · 2779.9B | 1524.5 | ~1674 GB |
| 30 | GPT 5.4 (Mar 05, 2026, medium) · proprietary | 1520.7 | API only |
| 31 | Grok 4.6 (xhigh) · proprietary | 1508.1 | API only |
| 32 | Claude Sonnet 5 (high) · proprietary | 1463.1 | API only |
| 33 | Claude Opus 4.8 None · proprietary | 1411.8 | API only |
| 34 | DeepSeek V4 Pro 0813 · 1650.5B | 1403.2 | ~991 GB |
| 35 | Gemini 3 Flash Preview · proprietary | 1367.2 | API only |
| 36 | Claude Sonnet 4.6 (medium) · proprietary | 1327.3 | API only |
| 37 | Claude Opus 4.7 · proprietary | 1323.0 | API only |
| 38 | GLM 5.3 · 753.3B | 1317.4 | ~456 GB |
| 39 | Grok 4.5 (high) · proprietary | 1309.3 | API only |
| 40 | DeepSeek V4 Flash 0731 · 304.2B | 1306.1 | ~183 GB |
| 41 | GPT 5.2 Codex · proprietary | 1299.9 | API only |
| 42 | GPT 5.2 (Dec 11, 2025, high) · proprietary | 1293.5 | API only |
| 43 | Gemini 3.8 Flash (high) · proprietary | 1269.7 | API only |
| 44 | GPT 5.2 (Dec 11, 2025, medium) · proprietary | 1249.8 | API only |
| 45 | GPT 5.1 Codex · proprietary | 1244.9 | API only |
| 46 | MiMo V2.6 Flash RL · 309B | 1244.2 | ~186 GB |
| 47 | GPT 5.1 Codex Max (xhigh) · proprietary | 1228.3 | API only |
| 48 | Mistral Large 4 1T A52B · 1000B | 1213.2 | ~604 GB |
| 49 | GPT 5.1 Codex Max · proprietary | 1208.8 | API only |
| 50 | GPT 5.1 (Nov 13, 2025, high) · proprietary | 1192.2 | API only |
| 51 | Qwen3.7 Max · proprietary | 1189.4 | API only |
| 52 | GPT 5.4 Mini (Mar 17, 2026, high) · proprietary | 1188.6 | API only |
| 53 | Gemini 3 Pro Preview · proprietary | 1176.8 | API only |
| 54 | GPT 5 (Aug 07, 2025, high) · proprietary | 1162.5 | API only |
| 55 | Gemini 3.1 Pro Preview · proprietary | 1160.6 | API only |
| 56 | MiMo V2.6 Pro RL · 1020B | 1158.1 | ~613 GB |
| 57 | Grok 4.20 · proprietary | 1150.3 | API only |
| 58 | GPT 5.5 None · proprietary | 1127.6 | API only |
| 59 | Beam 1.0 501B A23B · 501.3B | 1097.5 | ~301 GB |
| 60 | Kimi K2.6 · 1026.9B | 1092.7 | ~620 GB |
| 61 | DeepSeek V4.1 Flash · 763.2B | 1092.3 | ~458 GB |
| 62 | GPT 5.4 2026 03.05 None · proprietary | 1086.0 | API only |
| 63 | Gemini 3.1 Pro Preview (low) · proprietary | 1054.8 | API only |
| 64 | GLM 5.2 · 753.3B | 1047.0 | ~456 GB |
| 65 | Claude Opus 4.5 (Nov 01, 2025, 16K) · proprietary | 1025.4 | API only |
| 66 | DeepSeek V4 Pro · 1598.8B | 1006.1 | ~960 GB |
| 67 | GPT 5.4 Nano (Mar 17, 2026, high) · proprietary | 1004.5 | API only |
| 68 | Claude Opus 4.6 · proprietary | 996.5 | API only |
| 69 | Gemini 3 Pro Preview (low) · proprietary | 988.2 | API only |
| 70 | Inkling · 952.4B | 946.0 | ~572 GB |
| 71 | Grok 4.3 (unspecified) · proprietary | 944.2 | API only |
| 72 | O3 (Apr 16, 2025, high) · proprietary | 933.5 | API only |
| 73 | Gemma 4 26B A4B IT · 25.8B | 927.2 | ~16 GB |
| 74 | Gemma 4 31B IT · 31.3B | 925.5 | ~20 GB |
| 75 | Gemini 3.5 Flash (high) · proprietary | 911.0 | API only |
| 76 | Gemini 3.7 Flash (high) · proprietary | 904.3 | API only |
| 77 | MiMo V2.5 Pro · 1023.2B | 899.8 | ~614 GB |
| 78 | GLM 5.1 · 753.9B | 887.1 | ~457 GB |
| 79 | Kimi K2.7 Code · 1026.9B | 886.2 | ~620 GB |
| 80 | O4 Mini (Apr 16, 2025, high) · proprietary | 826.2 | API only |
| 81 | Kimi K2.5 · 1026.9B | 821.6 | ~620 GB |
| 82 | GPT 5 (Aug 07, 2025, minimal) · proprietary | 807.6 | API only |
| 83 | DeepSeek R1 0528 · 684.5B | 804.1 | ~415 GB |
| 84 | GPT 5 Mini (Aug 07, 2025, high) · proprietary | 799.8 | API only |
| 85 | Gemini 3.1 Flash Lite · proprietary | 797.7 | API only |
| 86 | Claude Sonnet 4.5 (Sep 29, 2025, 32K) · proprietary | 796.1 | API only |
| 87 | Mercury 2 · proprietary | 785.6 | API only |
| 88 | Gemini 2.5 Pro (32K) · proprietary | 785.5 | API only |
| 89 | Mimo v2 Pro · proprietary | 785.2 | API only |
| 90 | GLM 5 · 753.9B | 765.6 | ~457 GB |
| 91 | Gemini 3.5 Flash Lite (high) · proprietary | 765.3 | API only |
| 92 | Mistral Medium 3.5 128B · 127.7B | 764.0 | ~78 GB |
| 93 | DeepSeek V3.1 Terminus · 684.5B | 745.2 | ~415 GB |
| 94 | MiMo v2 Flash · 309.8B | 738.0 | ~186 GB |
| 95 | GPT 5 Nano (Aug 07, 2025, high) · proprietary | 718.7 | API only |
| 96 | Gemini 3.6 Flash (high) · proprietary | 715.5 | API only |
| 97 | Step 3.7 Flash · 201.4B | 694.1 | ~133 GB |
| 98 | DeepSeek V4 Flash · 290.9B | 678.2 | ~175 GB |
| 99 | Claude Opus 4.1 (Aug 05, 2025, 16K) · proprietary | 674.8 | API only |
| 100 | Qwen3.6 Plus · proprietary | 670.1 | API only |
| 101 | Gemini 2.5 Flash · proprietary | 661.9 | API only |
| 102 | Claude Sonnet 4 (May 14, 2025, 32K) · proprietary | 655.4 | API only |
| 103 | Claude Haiku 4.5 (Oct 01, 2025, 32K) · proprietary | 653.5 | API only |
| 104 | MiniMax M3 · 427.0B | 640.0 | ~257 GB |
| 105 | MiniMax M2.1 · 228.7B | 623.8 | ~138 GB |
| 106 | Qwen3.5 397B A17B · 403.4B | 621.9 | ~242 GB |
| 107 | MiniMax M2.5 · 228.7B | 618.2 | ~138 GB |
| 108 | Qwen3.5 Plus · proprietary | 608.9 | API only |
| 109 | MiniMax M2.7 · 228.7B | 599.3 | ~138 GB |
| 110 | Kimi K2 Thinking · 1026.4B | 597.5 | ~620 GB |
| 111 | Grok Code Fast 1 · proprietary | 587.7 | API only |
| 112 | GPT OSS 120B · 116.8B | 575.6 | ~71 GB |
| 113 | GPT OSS 20B · 20.9B | 566.0 | ~13 GB |
| 114 | GPT 4.1 (Apr 14, 2025) · proprietary | 558.1 | API only |
| 115 | MiMo V2.5 · 310.8B | 514.0 | ~187 GB |
| 116 | Mistral Small 4 119B 2603 · 119.4B | 497.6 | ~73 GB |
| 117 | Qwen3 Coder 480B A35B Instruct · 480.2B | 461.4 | ~289 GB |
| 118 | Qwen3 Coder Plus · proprietary | 456.5 | API only |
| 119 | Ring 2.6 1T · 1025.4B | 432.6 | ~2057 GB |
| 120 | GLM 4.7 · 358.3B | 399.5 | ~216 GB |
| 121 | Grok 4.1 Fast Reasoning · proprietary | 394.9 | API only |
| 122 | Qwen3 Max (Sep 23, 2025) · proprietary | 370.4 | API only |
| 123 | Qwen3.5 27B · 27.8B | 349.4 | ~17 GB |
| 124 | GLM 4.5 · 358.3B | 344.8 | ~216 GB |
| 125 | GLM 4.6 · 356.8B | 340.8 | ~215 GB |
| 126 | Qwen3.6 Flash · proprietary | 326.4 | API only |
| 127 | Gemini 2.5 Flash Lite Preview Thinking (Jun 17) · proprietary | 325.9 | API only |
| 128 | GLM 5.3 Flash · 321.3B | 303.6 | ~195 GB |
| 129 | Mercury 2.5 (high) · proprietary | 301.6 | API only |
| 130 | Kimi K2 Instruct 0905 · 1026.4B | 267.1 | ~620 GB |
| 131 | Qwen3.5 Flash · proprietary | 265.9 | API only |
| 132 | Mistral Large 3 675B Instruct 2512 · 675B | 264.7 | ~446 GB |
| 133 | Amazon.nova Lite v1:0 · proprietary | 236.3 | API only |
| 134 | Qwen3.5 35B A3B · 36.0B | 221.8 | ~22 GB |
| 135 | NVIDIA Nemotron 3 Super 120B A12B BF16 · 123.6B | 213.9 | ~75 GB |
| 136 | Mistral Medium 2508 · proprietary | 210.2 | API only |
| 137 | Llama 4 Maverick 17B 128E Instruct · 401.6B | 173.0 | ~265 GB |
| 138 | Codestral 2508 · proprietary | 137.8 | API only |
Score vs model size
Which models give the most quality for their size — the ones worth running locally.
- GPT OSS 20B, 21B, score 566.0 — on the efficiency frontier (best score at its size or smaller).
- Gemma 4 26B A4B IT, 26B, score 927.2 — on the efficiency frontier (best score at its size or smaller).
- DeepSeek V4 Flash 0731, 304B, score 1306.1 — on the efficiency frontier (best score at its size or smaller).
- GLM 5.3, 753B, score 1317.4 — on the efficiency frontier (best score at its size or smaller).
- DeepSeek V4.1 Flash 1017, 763B, score 1622.0 — on the efficiency frontier (best score at its size or smaller).
- Kimi K3.1 Pro, 2.8T, score 1932.8 — on the efficiency frontier (best score at its size or smaller).
ALE-Bench: frequently asked questions
- What is the best open LLM on ALE-Bench?
- Kimi K3.1 Pro is the top open model on ALE-Bench, scoring 1932.8. Among all models tested — including proprietary ones — it ranks #14. The top model overall is GPT 6 Astra Max (OpenAI) at 2951.3.
- What's the best ALE-Bench model you can run on a 24 GB GPU?
- Gemma 4 26B A4B IT is the highest-scoring open model that fits in 24 GB at 4-bit quantization (~16 GB), scoring 927.2 on ALE-Bench.
- Can open models match proprietary models on ALE-Bench?
- Not quite on ALE-Bench: the strongest proprietary model (GPT 6 Astra Max) scores 2951.3, ahead of the best open model (Kimi K3.1 Pro) at 1932.8 — 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.