Coding
WeirdML Leaderboard
WeirdML asks models to write machine-learning code for small, deliberately unusual datasets, then runs the code and scores the accuracy it actually achieves. Built by Håvard Tveit Ihle, it rewards practical problem-solving over recognizing a familiar task.
Source: epoch47 open models ranked+115 proprietaryData through Sep 2026
All models ranked on WeirdML
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 Promax · proprietary | 93.6% | API only |
| 2 | GPT 6 Astra Max · proprietary | 93.3% | API only |
| 3 | Claude Fable 5.1 Max · proprietary | 92.9% | API only |
| 4 | GPT 6 Astra (high) · proprietary | 92.9% | API only |
| 5 | Claude Fable 5.1 (high) · proprietary | 92.3% | API only |
| 6 | Claude Fable 5 Max · proprietary | 91.9% | API only |
| 7 | Claude Opus 5 Max · proprietary | 91.8% | API only |
| 8 | Claude Opus 5 (high) · proprietary | 91.6% | API only |
| 9 | GPT 5.6 Sol Promax · proprietary | 89.4% | API only |
| 10 | GPT 5.6 Sol (high) · proprietary | 88.8% | API only |
| 11 | Claude Fable 5 (high) · proprietary | 87.8% | API only |
| 12 | GPT 5.6 Sol Max · proprietary | 87.0% | API only |
| 13 | Claude Opus 5 · proprietary | 86.3% | API only |
| 14 | GPT 5.5 (xhigh) · proprietary | 84.9% | API only |
| 15 | Gemini 3.8 Flash (high) · proprietary | 84.8% | API only |
| 16 | GPT 5.5 (high) · proprietary | 83.9% | API only |
| 17 | Claude Opus 4.8 (xhigh) · proprietary | 82.9% | API only |
| 18 | Kimi K3 · 2779.9B | 82.6% | ~1674 GB |
| 19 | GPT 5.3 Codex · proprietary | 79.3% | API only |
| 20 | GPT 5.6 Terra (high) · proprietary | 78.3% | API only |
| 21 | Claude Opus 4.6 (high) · proprietary | 78.0% | API only |
| 22 | Claude Opus 4.6 (unspecified) · proprietary | 77.9% | API only |
| 23 | GPT 5.3 Codex (xhigh) · proprietary | 77.9% | API only |
| 24 | GPT 5.4 (Mar 05, 2026, xhigh) · proprietary | 77.7% | API only |
| 25 | Claude Opus 4.7 (high) · proprietary | 76.4% | API only |
| 26 | Claude Opus 4.7 · proprietary | 76.4% | API only |
| 27 | Claude Opus 4.7 (unspecified) · proprietary | 76.4% | API only |
| 28 | Claude Opus 4.8 (medium) · proprietary | 76.0% | API only |
| 29 | Claude Opus 4.7 Max · proprietary | 75.4% | API only |
| 30 | GLM 5.3 · 753.3B | 75.4% | ~456 GB |
| 31 | GPT 5.2 (Dec 11, 2025, xhigh) · proprietary | 72.2% | API only |
| 32 | Gemini 3.1 Pro Preview · proprietary | 72.1% | API only |
| 33 | Claude Opus 4.8 None · proprietary | 70.5% | API only |
| 34 | GLM 5.2 · 753.3B | 70.1% | ~456 GB |
| 35 | Gemini 3 Pro Preview · proprietary | 69.9% | API only |
| 36 | Claude Sonnet 5 (high) · proprietary | 68.8% | API only |
| 37 | Grok 4.6 (high) · proprietary | 67.3% | API only |
| 38 | GPT 5.5 None · proprietary | 67.2% | API only |
| 39 | DeepSeek V4 Pro 0813 · 1650.5B | 66.2% | ~991 GB |
| 40 | Claude Sonnet 4.6 (medium) · proprietary | 66.1% | API only |
| 41 | Claude Opus 4.6 · proprietary | 65.9% | API only |
| 42 | Claude Opus 4.5 (Nov 01, 2025, 16K) · proprietary | 63.7% | API only |
| 43 | GPT 5.2 (Dec 11, 2025, medium) · proprietary | 63.4% | API only |
| 44 | DeepSeek V4 Flash 0731 · 304.2B | 63.0% | ~183 GB |
| 45 | Gemini 3.5 Flash (high) · proprietary | 62.6% | API only |
| 46 | Gemini 3 Flash Preview · proprietary | 61.6% | API only |
| 47 | GPT 5.6 Luna (high) · proprietary | 60.9% | API only |
| 48 | GPT 5.1 (Nov 13, 2025, high) · proprietary | 60.8% | API only |
| 49 | GPT 5 (Aug 07, 2025, high) · proprietary | 60.7% | API only |
| 50 | GPT 5 Pro (Oct 06, 2025, high) · proprietary | 60.4% | API only |
| 51 | GPT 5.4 Mini (Mar 17, 2026, high) · proprietary | 60.3% | API only |
| 52 | Muse Spark 1.2 (xhigh) · proprietary | 60.3% | API only |
| 53 | O3 Pro (Jun 10, 2025, high) · proprietary | 58.2% | API only |
| 54 | GPT 5.4 2026 03.05 None · proprietary | 57.4% | API only |
| 55 | GPT 5.4 Pro 2026 03.05 None · proprietary | 57.4% | API only |
| 56 | GLM 5.1 · 753.9B | 57.1% | ~457 GB |
| 57 | Gemini 3.6 Flash (high) · proprietary | 56.1% | API only |
| 58 | Kimi K2.6 · 1026.9B | 55.9% | ~620 GB |
| 59 | GPT 5 Codex · proprietary | 54.5% | API only |
| 60 | GPT 5 Codex (high) · proprietary | 54.5% | API only |
| 61 | Kimi K2.7 Code · 1026.9B | 54.1% | ~620 GB |
| 62 | Gemini 2.5 Pro (16K) · proprietary | 54.0% | API only |
| 63 | GPT 5 Mini (Aug 07, 2025, high) · proprietary | 52.7% | API only |
| 64 | O4 Mini (Apr 16, 2025, high) · proprietary | 52.6% | API only |
| 65 | O3 (Apr 16, 2025, high) · proprietary | 52.4% | API only |
| 66 | Gemma 4 31B IT · 31.3B | 52.3% | ~20 GB |
| 67 | Grok 4.20 · proprietary | 52.3% | API only |
| 68 | Gemini 3.1 Flash Lite · proprietary | 52.2% | API only |
| 69 | Grok 4.3 (unspecified) · proprietary | 49.9% | API only |
| 70 | GPT 5.2 (Dec 11, 2025, low) · proprietary | 49.6% | API only |
| 71 | GPT 5.2 2025 12.11 None · proprietary | 49.6% | API only |
| 72 | GPT 5.4 Nano (Mar 17, 2026, high) · proprietary | 49.2% | API only |
| 73 | DeepSeek V4 Pro · 1598.8B | 48.9% | ~960 GB |
| 74 | GLM 5 · 753.9B | 48.2% | ~457 GB |
| 75 | GPT OSS 120B · 116.8B | 48.2% | ~71 GB |
| 76 | Claude Sonnet 4.5 (Sep 29, 2025, 16K) · proprietary | 47.7% | API only |
| 77 | O1 Preview (Sep 12, 2024) · proprietary | 47.6% | API only |
| 78 | DeepSeek V3.2 Speciale · 685.4B | 46.7% | ~415 GB |
| 79 | Claude Sonnet 4.5 (Sep 29, 2025) · proprietary | 46.7% | API only |
| 80 | Grok 4.5 (unspecified) · proprietary | 46.4% | API only |
| 81 | Claude Sonnet 4 (May 14, 2025, 16K) · proprietary | 46.1% | API only |
| 82 | O1 (Dec 17, 2024, high) · proprietary | 46.1% | API only |
| 83 | Claude Opus 4.1 (Aug 05, 2025, 16K) · proprietary | 45.9% | API only |
| 84 | Grok 4 (Jul 09) · proprietary | 45.7% | API only |
| 85 | DeepSeek V4 Flash · 290.9B | 45.6% | ~175 GB |
| 86 | Kimi K2.5 · 1026.9B | 45.6% | ~620 GB |
| 87 | Claude Haiku 4.5 (Oct 01, 2025) · proprietary | 45.4% | API only |
| 88 | Claude Haiku 4.5 (Oct 01, 2025, 16K) · proprietary | 44.1% | API only |
| 89 | Claude Sonnet 4 (May 14, 2025) · proprietary | 43.9% | API only |
| 90 | Claude Opus 4 (May 14, 2025, 16K) · proprietary | 43.7% | API only |
| 91 | Mistral Medium 3.5 128B · 127.7B | 43.7% | ~78 GB |
| 92 | O3 Mini (Jan 31, 2025, high) · proprietary | 43.7% | API only |
| 93 | NVIDIA Nemotron 3 Ultra 550B A55B BF16 · 560.5B | 43.5% | ~370 GB |
| 94 | Mercury 2 · proprietary | 43.2% | API only |
| 95 | Grok 4 Fast · proprietary | 42.9% | API only |
| 96 | Kimi K2 Thinking · 1026.4B | 42.8% | ~620 GB |
| 97 | Grok 3 Mini (high) · proprietary | 42.6% | API only |
| 98 | Grok 3 Mini Beta (high) · proprietary | 42.6% | API only |
| 99 | Gemini 2.5 Flash Preview (Sep 2025, 16K) · proprietary | 41.9% | API only |
| 100 | DeepSeek R1 0528 · 684.5B | 41.6% | ~415 GB |
| 101 | Qwen3 Coder 480B A35B Instruct · 480.2B | 41.2% | ~289 GB |
| 102 | Qwen3 235B A22B Thinking 2507 · 235.1B | 41.0% | ~142 GB |
| 103 | Gemini 2.5 Flash Preview (16K thinking) (Apr 17) · proprietary | 40.9% | API only |
| 104 | Gemini 2.5 Flash Preview (May 20, 16K) · proprietary | 40.9% | API only |
| 105 | GPT OSS 20B · 20.9B | 40.9% | ~13 GB |
| 106 | GLM 4.5 · 358.3B | 40.6% | ~216 GB |
| 107 | Claude 3.5 Sonnet (Oct 22, 2024) · proprietary | 40.0% | API only |
| 108 | GPT 5 Chat · proprietary | 39.8% | API only |
| 109 | Qwen3.5 27B · 27.8B | 39.5% | ~17 GB |
| 110 | DeepSeek V3.2 Exp · 685.4B | 39.5% | ~415 GB |
| 111 | GPT 4.5 Preview (Feb 27, 2025) · proprietary | 39.4% | API only |
| 112 | Kimi K2 Instruct · 1026.4B | 39.4% | ~620 GB |
| 113 | GPT 4.1 (Apr 14, 2025) · proprietary | 39.0% | API only |
| 114 | Gemini 3.5 Flash Lite (high) · proprietary | 39.0% | API only |
| 115 | Qwen3 235B A22B Instruct 2507 · 235.1B | 38.7% | ~142 GB |
| 116 | DeepSeek V3.1 · 684.5B | 38.4% | ~415 GB |
| 117 | GPT 5 Nano (Aug 07, 2025, high) · proprietary | 38.1% | API only |
| 118 | NVIDIA Nemotron 3 Super 120B A12B BF16 · 123.6B | 38.0% | ~75 GB |
| 119 | GPT 5.4 Nano 2026 03.17 None · proprietary | 38.0% | API only |
| 120 | GPT 5.4 Mini 2026 03.17 None · proprietary | 37.9% | API only |
| 121 | GPT 4.1 Mini (Apr 14, 2025) · proprietary | 37.6% | API only |
| 122 | Qwen3 235B A22B · 235.1B | 37.3% | ~142 GB |
| 123 | Grok 3 · proprietary | 37.2% | API only |
| 124 | MiniMax M2.7 · 228.7B | 37.0% | ~138 GB |
| 125 | Kimi K2 Instruct 0905 · 1026.4B | 36.7% | ~620 GB |
| 126 | DeepSeek R1 · 684.5B | 36.5% | ~415 GB |
| 127 | O1 Mini (Sep 12, 2024, medium) · proprietary | 36.3% | API only |
| 128 | DeepSeek v3 0324 · 684.5B | 36.1% | ~415 GB |
| 129 | Gemini 2.5 Flash Lite Preview (Jun 17, 16K) · proprietary | 35.2% | API only |
| 130 | Gemini 2.5 Flash Lite Preview Thinking (Jun 17) · proprietary | 35.2% | API only |
| 131 | Gemma 4 26B A4B IT · 25.8B | 35.2% | ~16 GB |
| 132 | Grok Code Fast 1 · proprietary | 35.1% | API only |
| 133 | Qwen3.6 35B A3B · 36.0B | 34.5% | ~22 GB |
| 134 | Qwen3 Coder Next · 79.7B | 34.4% | ~48 GB |
| 135 | Mistral Medium 2508 · proprietary | 33.1% | API only |
| 136 | Inkling · 952.4B | 32.3% | ~572 GB |
| 137 | Claude 3.5 Sonnet (Jun 20, 2024) · proprietary | 31.0% | API only |
| 138 | Claude 3.5 Haiku (Oct 22, 2024) · proprietary | 30.7% | API only |
| 139 | Qwen3 30B A3B · 30.5B | 29.8% | ~19 GB |
| 140 | GPT 5 Nano (Aug 07, 2025, low) · proprietary | 25.9% | API only |
| 141 | Gemini 2.0 Flash 001 · proprietary | 25.8% | API only |
| 142 | GPT 4o (Nov 20, 2024) · proprietary | 25.1% | API only |
| 143 | Gemini 1.5 Flash 002 · proprietary | 24.9% | API only |
| 144 | Llama 4 Maverick 17B 128E Instruct · 401.6B | 24.5% | ~265 GB |
| 145 | Grok 2 (Dec 12) · proprietary | 22.2% | API only |
| 146 | Gemini 1.5 Pro 002 · proprietary | 22.2% | API only |
| 147 | Llama 3.1 405B Instruct · 405.9B | 21.4% | ~268 GB |
| 148 | Claude 3 Opus (Feb 29, 2024) · proprietary | 19.2% | API only |
| 149 | GPT 4.1 Nano (Apr 14, 2025) · proprietary | 19.0% | API only |
| 150 | GPT 4 Turbo (Apr 09, 2024) · proprietary | 18.0% | API only |
| 151 | Qwen2.5 72B Instruct · 72.7B | 16.0% | ~45 GB |
| 152 | Llama 3.3 70B Instruct · 70.6B | 14.4% | ~47 GB |
| 153 | GPT 4 (Jun 13) · proprietary | 12.4% | API only |
| 154 | GPT 4o Mini (Jul 18, 2024) · proprietary | 11.8% | API only |
| 155 | Qwen2 72B Instruct · 72.7B | 11.3% | ~45 GB |
| 156 | Claude 3 Sonnet (Feb 29, 2024) · proprietary | 10.2% | API only |
| 157 | Claude 3 Haiku (Mar 07, 2024) · proprietary | 9.8% | API only |
| 158 | Llama 3.1 70B Instruct · 70.6B | 9.0% | ~47 GB |
| 159 | Claude 2.1 · proprietary | 7.1% | API only |
| 160 | GPT 3.5 Turbo (Jan 25) · proprietary | 3.5% | API only |
| 161 | Mixtral 8x22B Instruct v0.1 · 140.6B | 3.2% | ~85 GB |
| 162 | Llama 3.1 8B Instruct · 8.0B | 1.7% | ~5.3 GB |
Score vs model size
Which models give the most quality for their size — the ones worth running locally.
- Llama 3.1 8B Instruct, 8B, score 1.7% — on the efficiency frontier (best score at its size or smaller).
- GPT OSS 20B, 21B, score 40.9% — on the efficiency frontier (best score at its size or smaller).
- Gemma 4 31B IT, 31B, score 52.3% — on the efficiency frontier (best score at its size or smaller).
- DeepSeek V4 Flash 0731, 304B, score 63.0% — on the efficiency frontier (best score at its size or smaller).
- GLM 5.3, 753B, score 75.4% — on the efficiency frontier (best score at its size or smaller).
- Kimi K3, 2.8T, score 82.6% — on the efficiency frontier (best score at its size or smaller).
WeirdML: frequently asked questions
- What is the best open LLM on WeirdML?
- Kimi K3 is the top open model on WeirdML, scoring 82.6%. Among all models tested — including proprietary ones — it ranks #18. The top model overall is GPT 6 Astra Promax (OpenAI) at 93.6%.
- What's the best WeirdML 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 (~20 GB), scoring 52.3% on WeirdML.
- What's the best WeirdML model you can run on a 12 GB GPU?
- Llama 3.1 8B Instruct is the highest-scoring open model that fits in 12 GB at 4-bit quantization (~5.3 GB), scoring 1.7% on WeirdML.
- Can open models match proprietary models on WeirdML?
- Not quite on WeirdML: the strongest proprietary model (GPT 6 Astra Promax) scores 93.6%, ahead of the best open model (Kimi K3) at 82.6% — 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.