Will the highest SciCode score score in Epoch AI's benchmark data reach at least 70.5 % for a model released on or before 2026-12-13?
deadline 2026-12-13
confidence medium
tier 1 evidence
23%PolySignal estimate
Resolution criterion
Resolves YES if, in a snapshot of Epoch AI's benchmark data (https://epoch.ai/data/benchmark_data.zip, file "scicode_external.csv", column "Score") taken on or after 2026-12-27, at least one model with a release date on or before 2026-12-13 has a value >= 70.5 %. Reference at creation: 66.9 % (claude-opus-5-5_max, 2026-09-22). Voided if Epoch AI revises the methodology so that values are no longer comparable, or retires the file. Data: Epoch AI, CC BY 4.0.
Verified starting state
Computed base rate: 17% of past 60-day windows in Epoch AI's benchmark data saw the running record rise by at least the required amount (threshold 0.705, current record 0.668981481481482). Data: Epoch AI, CC BY 4.0, epoch.ai/benchmarks. Overlapping windows: small effective sample.
What pushes it up
- The current record of 66.9% was established very recently (2026-09-22) by Claude Opus 5.5, indicating that the benchmark is being actively targeted by frontier models [1].
- A 3.6 percentage point improvement is consistent with the typical performance jumps seen between iterative releases of flagship AI models in scientific reasoning and coding tasks [2].
What pushes it down
- The historical base rate for achieving this specific score increase within a 60-day window is relatively low at 17% [1].
- SciCode is a specialized benchmark for scientific coding, where gains often become more difficult to achieve as models reach higher performance tiers [1].
- There are no specific announcements of upcoming model releases from major labs (OpenAI, Anthropic, Google) within the next 60 days that explicitly target this benchmark [1].
What would move this number most
The timing of the next major frontier model release (e.g., a successor to Claude 5.5 or GPT-5) before the December 13 deadline.
How this number is built
| Deadline | 2026-12-13 |
| Base rate | 17% |
| Best evidence | tier 1 · 1 article(s) used |
| Independent runs | 20 · 25 (spread 5 pts) |
| Confidence | medium |
| Evidence cap | not triggered |
| Last revised | 2026-10-14 07:57 |
| Model | gemini-3-flash-preview |
Sources consulted