METR report lifts curtain on Anthropic's internal 'Model 4' as capability gains hit ~2x historical rate
NinaDSchick · x · 2026-11-06
METR published a public report with rare details on Anthropic's internal models:
- Acceleration evidence: As of October 2026, Anthropic is likely improving model capabilities on well-specified tasks at 2x the rate of March 2024–October 2025 (high uncertainty; could be 1.3x–3x). This underpins Anthropic's 'dramatic acceleration of AI R&D' disclosure.
- Internal Model 4: finished training around September–October, scoring 178.6 on AECI (a different scale than prior system cards).
- Known scores: Opus 5.5 174.5, Mythos 5.1 172.5.
- METR notes AI cannot yet make highly significant fully autonomous contributions.
The chart shows AECI trend accelerating from +15/yr to +29/yr, with internal models (Model 2/3/4 series) consistently ahead of public releases.
More from Models
- Diffusion-Qwen3.8-27B: Qwen3 adapted into a masked diffusion LM with bidirectional attention — GabGarrett · 2026-11-06
- Researcher slams Inkling 2 Small time-series demo as a practically pointless task — mariofilhoml · 2026-11-06
- Inkling 2 Small jumps 28.5 points to become top US open-weight model on Vals Index — simonguozirui · 2026-11-06
- Small-model customizability called critical as Inkling 2 Small touts SOTA fine-tuning — clarejtbirch · 2026-11-06
- TensorFold 1.2.3 boosts Flash Next decode 21–23%; Apache-2.0 Ripple-65B Preview lands — AIFlow_ML · 2026-11-06
- Inkling-2-Small launches: a small, open-source agentic model with native vision — clarejtbirch · 2026-11-06