Math is a compute story, instruction-following a post-training one, says ex-Microsoft exec
MParakhin · x · 2026-10-11
Mikhail Parakhin (ex-Microsoft) compares two frontier models: Astra (especially Pro) beats Fable at math, while Fable is better at following instructions. He concludes math is largely a compute story, whereas instruction following comes down to post-training — yet he doubts Astra's first-rate team did sloppy tuning and asks what else might explain the gap.
More from Models
- Alexandr Wang launches Muse, touting adoption faster than ChatGPT, Grok and Gemini — alexandr_wang · 2026-10-11
- repligate: fable 5.1 is the 'horniest' model since Claude 3 Opus — repligate · 2026-10-11
- Leaks claim OpenAI still has two pretrain runs dwarfing Google's astra — iruletheworldmo · 2026-10-11
- 75-test coding benchmark pits RTX 4090 vs Strix Halo vs M5 Max for local LLMs — julianharris · 2026-10-11
- Anthropic reportedly asked AWS to urgently pull Sonnet 3.5, 3.6 and 3.7 from Bedrock — repligate · 2026-10-11
- Benchmark audits find ~30% of SWE-Bench Pro tasks broken; SciCode and Terminal-Bench flawed too — geoffwolfe · 2026-10-11