Track record
Every forecast is logged before launch and graded afterwards. None have resolved yet. The first results come when Mistral Large 4 weights and Flint land, or their windows close.
| Model | Logged | Window | Chance | Result |
|---|---|---|---|---|
| Mistral Large 4 (free download) | Oct 26, 2026 | Oct 26, 2026 – Nov 20, 2026 | 75% | Open |
| Qwen 4 | Oct 26, 2026 | Oct 27, 2026 – Jan 15, 2027 | 81% | Open |
| GPT-6.1 Luna | Oct 26, 2026 | Nov 2, 2026 – Nov 30, 2026 | 30% | Open |
| Claude Fable 5.2 | Oct 26, 2026 | Oct 28, 2026 – Feb 15, 2027 | 35% | Open |
| DeepSeek V4.1 Pro | Oct 26, 2026 | Nov 1, 2026 – Feb 28, 2027 | 55% | Open |
| Grok 5 | Oct 26, 2026 | Nov 15, 2026 – Apr 30, 2027 | 45% | Open |
| Muse Spark (free download) | Oct 26, 2026 | Dec 1, 2026 – Aug 31, 2027 | 43% | Open |
| Claude 6 | Oct 26, 2026 | Feb 1, 2027 – Oct 31, 2027 | 40% | Open |
| Llama 5 | Oct 26, 2026 | Jan 1, 2027 – Dec 31, 2027 | 10% | Open |
| GPT-7 | Oct 26, 2026 | Mar 1, 2027 – Dec 31, 2027 | 43% | Open |
How we grade
When a window closes, the outcome is 1 if the model shipped inside it and 0 if not. We report the Brier score: the average of (chance − outcome)². Zero is perfect, and always guessing 50% scores 0.25.