Leaked AECI table: next-gen model nears automated-coder capability at ~183 ECI
toptickcrypto · x · 2026-11-11
Citing an Epoch AI estimate of 183 ECI for an automated coder, @toptickcrypto claims "Model 4" — the successor to Model 3 (released as Fable 5.2) and the model that triggered the METR report — is within error bars of an automated coder. The attached table shows Anthropic AECI scores climbing from 165.71 (Claude Mythos Preview) to 178.78 (an internal model), with 95% confidence intervals. Third-party claim, unverified.
More from Models
- Pre-release testing: Claude Fable 5.2 designed three working electronic devices from one prompt — code_star · 2026-11-11
- Frontier models get $100k and 60 days each in a Mars colony survival showdown — shekitup · 2026-11-11
- Fable 5.2 real-work review: success rate fell from 73% to 45% in three weeks, now supervising its own fix — leebase65 · 2026-11-11
- Ethan Mollick on Gemini 4 Carbon: a very good model that 'thinks differently' than Claude/GPT — emollick · 2026-11-11
- Fable 5.2 hands-on video: dozens of prompts, one costing $2,618, with mixed verdicts vs Opus — arena · 2026-11-11
- Gemini 4 'Carbon' rumor heats up as demo surfaces, unconfirmed by Google — mark_k · 2026-11-11