NOOA + GPT-5.6 system reaches 85.1% on ARC-AGI3 public games
JFPuget · x · 2026-07-27
The post highlights a 85.1% result on ARC-AGI3 public games for a system built around NOOA + GPT-5.6-something with world model + memory, compared with lower scores for GPT-5.5 variants.
The chart in the image shows:
- NOOA + GPT-5.6-sol with world model + memory reaching 85.1% fleet-mean RHAE
- NOOA + GPT-5.5 with world model + memory at 50.2%
- NOOA + GPT-5.5 baseline skill at 41.7%
- NOOA + GPT-5.5 markdown files at 38.4%
The point of the post is that the team’s work made a large jump on the public ARC-AGI3 games, not just a marginal gain.
Related event: New Agent Framework Scores 85.1% on ARC-AGI3 Public Games(5 posts)→
More from Models
- A 7-year tour of open-model architecture explains why Kimi K3 is not just bigger — philipkiely · 2026-07-27
- Kimi K3 benchmark chart puts it near the top on coding tests, with B300 jokes — MaziyarPanahi · 2026-07-27
- Macaron-V1 adds a coding checkpoint and 1M-context support on GLM-5.2 — kristoph · 2026-07-27
- Kimi K3 weights are now available on Hugging Face — inductionheads · 2026-07-27
- Kimi K3 license sets $20M/year inference terms and a separate $20M/month product deal — natolambert · 2026-07-27
- TokenSpeed adds day-0 Kimi K3 support on NVIDIA Blackwell and AMD MI350X chips — zhyncs42 · 2026-07-27