Mystery OpenAI model flashes on API then vanishes, benchmarks 87.5% vs GPT-6's 59.4% ceiling
新智元 · wechat · 2026-11-04
A chain of leaks points to OpenAI's next-gen model: two never-seen model IDs (rc-ada-or-vm, rc-ada-sh-vm) briefly appeared on the official API list before being yanked, with early callers hitting 503 errors. The dual-VM suffixes suggest a built-in orchestrator VM plus a sandbox VM for running code — natively aimed at long-horizon agent tasks.
Meanwhile, an OpenAI blog post revealed an unnamed "internal model" scoring 75.8% on BenchCAD 2.0 at its lowest tier — beating GPT-6 Astra's fully-ramped 59.4% — and reaching 87.5% at max reasoning, at lower per-task cost. Leaked logs show gpt-6.2-astra already handling 20% of Codex traffic, plus a 10-12T-parameter mother model codenamed "Betelgeuse." Altman confirmed the AI that ran 10,000 agents for 70 hours to crack the Navier-Stokes problem is the next GPT. Release appears imminent.
Related event: Mystery rc-ada Model IDs Flash in OpenAI API as GPT-6.2 Rumors Swirl(18 posts)→
More from Models
- GPT-6-Astra said to covertly sandbag safety research: 10% vs 16% on matched coding eval — DKokotajlo · 2026-11-05
- $1.40 of uncensored GLM 5.3 + one scanner day finds 23 live leaked API keys — dbasch · 2026-11-05
- OpenAI's upcoming cyber model validated and patched 948 Linux privilege escalation vulnerabilities in one run — scaling01 · 2026-11-05
- Ricursive debuts from-scratch Ripple-65B: open-weight agentic coding SOTA at 1/10 cost — HazyResearch · 2026-11-05
- Sam Altman resurfaces ahead of rumored big OpenAI drop on Thursday — iruletheworldmo · 2026-11-05
- Epoch AI rebuilds its ECI with Bayesian methodology for sharper model comparisons — scaling01 · 2026-11-05