Russian startup Mostik bridges LLM hidden states, cutting cost to 1/20
机器之心 · wechat · 2026-09-06
Mostik, a little-known Russian startup with 12 PhDs and Fields medalist Stanislav Smirnov, briefly topped the ARC-AGI-3 Kaggle leaderboard after just four months, keeping its competition system secret.
Core idea: models exchange hidden states directly
- Today's multi-agent systems pass text, but a model outputs only 17 bits per token while its internal state holds 2MB — like thinking in a thousand dimensions and speaking through a keyhole.
- Mostik trains a "bridge" (the only trained component; both models stay frozen) that translates GLM-5.2's (753B params) hidden states into Qwen-3.5 (4B params), letting the small model handle generation.
Results
- The hybrid system costs 1/20 of running GLM alone, gains 25% accuracy, closes 50% of the capability gap, and up to 2x on hard tasks.
- Hidden-state handoff beats text handoff by up to 10 points everywhere tested, with the advantage largest the earlier the large model hands off.
Extensions
- A new interpretability entry point: compare sender/translated/receiver states and run intervention experiments to test causal features.
- Next: training-time distillation with live teacher guidance, and "specialization plug-ins" — betting on model-to-model collaboration as a third scaling route.
All results are from Mostik's own disclosures; no external reproduction yet.
More from Models
- Local AI Goes Mainstream: TikTok Videos Have Ordinary People Asking to Run Qwen 27B at Home — StandardLovers · 2026-09-06
- OpenAI ships GPT-6 Astra, first model to hit Critical cyber threshold, at $10/$50 per token — btibor91 · 2026-09-06
- Instagram's AI detector mislabels real photos as 'AI Content', confusing users — emmanuelvivier · 2026-09-06
- Sam Altman reportedly apologizes for 'chaotic' GPT-6 Astra launch that locked out paying users — emmanuelvivier · 2026-09-06
- After 24 hours of testing, user calls Astra overhyped and says work usefulness peaked with Opus 4.8 — Scobleizer · 2026-09-06
- tokenbender on new model hype: trust shippers not yappers, verdict in 10-15 days — tokenbender · 2026-09-06