GPT-6 Sol and Claude Opus 5.2 leaks flood in days after AI slowdown pledge
APPSO · wechat · 2026-09-16
Days after Altman, Amodei and other AI leaders called for a slowdown, leaks about two new frontier models went viral:
GPT-6 Sol rumors (unconfirmed)
- OpenAI's Tibo teased "DevDay-scale" updates this week, with Altman's cryptic boost
- Test users report being quietly routed to GPT-6 Sol: 50%+ faster, 67% better token response times, a 2-hour task costing 3% of weekly quota—"a cheaper Astra"
- Users fingerprint the new model via knowledge cutoffs and packet-sniffing "juice" values
- Brockman also hinted at progress on "another Millennium Problem," widely guessed to be the Hodge Conjecture; separate leaks point to work on BSD
Claude Opus 5.2 internal testing (unconfirmed)
- Heavy Claude Code users report the model self-reviews and optimizes finished work, handles incomplete info without back-and-forth, and pushed through 8 complex tasks end-to-end
- In multi-agent tests, a suspected 5.2 sub-agent flagged another model's task allocation and demanded a re-split; tools fetched on-demand, only 60% context used after heavy work
- Visual tests: 2D line art expanded into detailed 3D scenes with multi-round refinement; a claude-opus-5-2 identifier spotted in Microsoft Foundry
The author's take: the slowdown pledge reads like psychological warfare—everyone wants the rival to brake first.
Related event: GPT-6 Sol and Opus 5.2 Leaks Emerge Days After Slowdown Plea(3 posts)→
More from Models
- Ant ships FP8, FP4 and INT4 quantized builds of Ling-3.0-flash-Fin for financial AI — nikola_mr64990 · 2026-09-16
- Meta's Muse hits 200,000+ posts on X in first 7 days after launch — ChrisUniverse · 2026-09-16
- Anthropic sends surprise usage credit and refund emails to users — Darpinian · 2026-09-16
- VisTW: a Traditional Chinese VLM benchmark for reading Taiwan — and an eval framework that caught a 36-point bug — piske_usagi · 2026-09-16
- Training Specialist Models Without Reasoning Trajectories: Latent Paths Drive Distillation Trade-offs — Yilei Tu · 2026-09-16
- Claude Opus 5's PRs proactively confess every mistake made while coding — repligate · 2026-09-16