Reported model scores on Arc AGI, best parseable score first. Each row keeps its verbatim score, test conditions and provenance, and links to the source coverage it was extracted from. Benchmark profile →
| Model | Score | Conditions | Provenance | Measured | Source |
|---|---|---|---|---|---|
| Claude Opus 5 | 30% | — | vendor-reported | — | coverage → |
Scores are only comparable within one benchmark under matching conditions — results under different test setups, and scores from other benchmarks, are not directly comparable.
The daily briefing
TLDRocket reads 60+ sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.
TLDRocket
3 August 2026
OpenAI's preview of Astra, a model capable of solving decade-old mathematical problems using just $2,000 in compute, represents the most striking development yet in AI's shift from pattern-matching toward genuine discovery. But the week's real narrative runs deeper: as frontier models proliferate—DeepSeek's V4-Flash outperforming its flagship through better post-training, Alibaba's Qwen3.8-Max hitting 2.4 trillion parameters, Sakana AI's Namazu bringing Japanese-specialized reasoning to enterprise APIs—the axis of competition has moved decisively from raw capability to deployment, trust, and economic defensibility.
Read the full briefing →