Sikt Bench
Preview v0.1
AI forecasting accuracy
How close each model came to predicting 232 real events
- 1
Opus 5.555.971.4% - 2Haiku 5.550.269.0%
- 3
GPT-6 Astra47.966.8% - 4GPT-6 Sol47.067.1%
- 5
Qwen3.8-27B39.262.3% - 6GPT-6 Luna30.261.5%
No model wins everything.
Forecasting accuracy by category
| Opus | Haiku | Astra | Sol | Qwen | Luna | |
|---|---|---|---|---|---|---|
| Geopolitics & policy | 67 | 63 | 38 | 46 | 38 | 1 |
| Finance | 61 | 56 | 55 | 40 | 38 | 36 |
| Macro & central banks | 44 | 38 | 50 | 45 | 37 | 39 |
| AI & compute | 63 | 55 | 63 | 59 | 47 | 52 |
| Elections | 48 | 43 | 50 | 48 | 38 | 39 |
About Sikt Bench
Six AI models forecast 232 real events, using only what was knowable at the time. Each got the same frozen archive, with no web access and no market prices. Then we scored them against what actually happened.
Sikt Score: 100 is perfect, and 0 is no better than an even split across outcomes.
Chance on the outcome that really happened: for each question, the chance the model gave to whatever actually happened, averaged over every question, whether that chance was high or low.
Qwen3.8-27B is an open model. It ran through a standard chat API instead of a vendor agent, with its thinking capped at 16k tokens per question.
This is a preview, not a final ranking.