Reasoning
ForecastBench Leaderboard
ForecastBench, from the Forecasting Research Institute, asks models to predict real-world future events, with questions that only resolve after the model's training cutoff so the answers can't have been memorised. It is scored as a difficulty-adjusted Brier index, where higher is better.
Source: epoch26 open models ranked+56 proprietaryData through Jul 2026
Open models ranked on ForecastBench
# shows rank among open models / rank overall (including proprietary).
| # | Model | Score |
|---|---|---|
| 1 / 12 | MiniMax M3 · 427.0B | 61.4 |
| 2 / 17 | Kimi K3 · 2779.9B | 61.1 |
| 3 / 18 | GLM 5 · 753.9B | 61.0 |
| 4 / 32 | Kimi K2 Instruct · 1026.4B | 60.2 |
| 5 / 35 | DeepSeek R1 · 684.5B | 60.0 |
| 6 / 37 | Llama 3.1 405B Instruct · 405.9B | 59.9 |
| 7 / 38 | Kimi K2 Instruct 0905 · 1026.5B | 59.8 |
| 8 / 39 | Qwen3 235B A22B · 235.1B | 59.7 |
| 9 / 45 | GLM 4.5 Air · 110.5B | 59.2 |
| 10 / 47 | DeepSeek v3 · 684.5B | 59.1 |
| 11 / 52 | Llama 3.3 70B Instruct · 70.6B | 58.6 |
| 12 / 53 | Meta Llama 3 8B Instruct · 8.0B | 58.6 |
| 13 / 57 | QwQ 32B Preview · 32.8B | 58.3 |
| 14 / 60 | DeepSeek V3.1 · 684.5B | 58.0 |
| 15 / 63 | Qwen1.5 110B Chat · 111.2B | 57.7 |
| 16 / 64 | Llama 4 Maverick 17B 128E Instruct · 401.6B | 57.5 |
| 17 / 65 | Llama 4 Scout 17B 16E Instruct · 108.6B | 57.5 |
| 18 / 66 | Qwen2.5 72B Instruct · 72.7B | 57.5 |
| 19 / 69 | Meta Llama 3 70B · 70.6B | 57.1 |
| 20 / 70 | Mistral Large Instruct 2407 · 122.6B | 57.1 |
| 21 / 72 | Mistral Large Instruct 2411 · 122.6B | 56.9 |
| 22 / 73 | Mixtral 8x22B Instruct v0.1 · 140.6B | 56.3 |
| 23 / 74 | Mixtral 8x7B Instruct v0.1 · 46.7B | 56.3 |
| 24 / 75 | DeepSeek V4 Pro · 1598.8B | 56.1 |
| 25 / 80 | Meta Llama 3 8B · 8.0B | 52.9 |
| 26 / 81 | Llama 2 70B Chat HF · 69.0B | 51.4 |
Score vs model size
Which models give the most quality for their size — the ones worth running locally.
- Meta Llama 3 8B Instruct, 8B, score 58.6 — on the efficiency frontier (best score at its size or smaller).
- GLM 4.5 Air, 110B, score 59.2 — on the efficiency frontier (best score at its size or smaller).
- Qwen3 235B A22B, 235B, score 59.7 — on the efficiency frontier (best score at its size or smaller).
- Llama 3.1 405B Instruct, 406B, score 59.9 — on the efficiency frontier (best score at its size or smaller).
- MiniMax M3, 427B, score 61.4 — on the efficiency frontier (best score at its size or smaller).
ForecastBench: frequently asked questions
- What is the best open LLM on ForecastBench?
- MiniMax M3 is the top open model on ForecastBench, scoring 61.4. Among all models tested — including proprietary ones — it ranks #9. The top model overall is O3 (Apr 16, 2025, unspecified) (OpenAI) at 62.5.
- What's the best ForecastBench model you can run on a 24 GB GPU?
- Meta Llama 3 8B Instruct is the highest-scoring open model that fits in 24 GB at 4-bit quantization (about 4 GB), scoring 58.6 on ForecastBench.
- What's the best ForecastBench model you can run on a 12 GB GPU?
- Meta Llama 3 8B Instruct is the highest-scoring open model that fits in 12 GB at 4-bit quantization (about 4 GB), scoring 58.6 on ForecastBench.
- Can open models match proprietary models on ForecastBench?
- Not quite on ForecastBench: the strongest proprietary model (O3 (Apr 16, 2025, unspecified)) scores 62.5, ahead of the best open model (MiniMax M3) at 61.4 — 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.