Open-weights Inkling climbs from last to top-3 on FutureSim via test-time learning
simonguozirui · x · 2026-11-18
On the FutureSim (simulated forecasting cycle) benchmark, the open-weights model Inkling shows the largest test-time improvement of any model: it starts at the bottom of the leaderboard but keeps climbing as the simulated forecasting cycle progresses, ultimately reaching top-3 — outperforming Muse Wonder and landing right next to GPT-6.1-Sol.
Researchers attribute this late-surging trajectory to very strong in-context learning — the model keeps improving from simulated feedback during inference — making it the best open-weights performer on the benchmark.
More from Models
- GPT-6.2 Astra beats Opus 5.5 across agent benchmarks at 3-6x lower cost — VraserX · 2026-11-18
- 3 weeks with OpenAI Astra 5.2: shines at delegating under messy stacked context — round · 2026-11-18
- OpenAI brings GPT-6.2 Astra and Sol to Codex and the API — reach_vb · 2026-11-18
- Perplexity hosts AMA on Decider V1.3 decision model, free on OpenRouter now — antoine_chaffin · 2026-11-18
- OpenAI ships GPT-6.2 Astra and Sol to ChatGPT Work, Codex, and API — pvncher · 2026-11-18
- OpenAI Astra for Law tops Harvey LAB-AA v1.1 legal benchmark at 20.8% — ArtificialAnlys · 2026-11-18