Open-weights CLM matches Jev's API surface, runs 4-13x faster with 75MB tunable heads
R_Duncan · reddit · 2026-09-24
CLM is an open-weights, self-hostable alternative to TypeSafe AI's Jev, implementing the same three decision primitives — Choice, Noul, and Score — with drop-in client compatibility.
Where CLM wins:
- Splits state and action heads; persistent action sets are embedded once and cached, yielding 4-13x speedups over the cloud-hosted Jev on browser-agent and gaming benchmarks (T-Rex, Super Mario)
- 75MB heads are fine-tunable on your own agent trajectories; fine-tuned CLM hits 87.6% on Terminal-Bench 2.1 and 81.6% on DeepSWE (zero-shot Jev scored 71% on DeepSWE)
- No API costs, fully self-hostable
Where Jev keeps the edge: stronger zero-shot out-of-domain accuracy (99.2% vs 95.2% on BFCL v4), 64K context vs CLM-8B's calibrated 2-8K, and internally calibrated absolute scoring. Net: no API features lost switching to CLM, only some niche zero-shot generalization.
More from coding & agent
- KAIST's WikiTrace: evidence-linked memory fixes agent "memory-induced degradation" — dair_ai · 2026-11-11
- Why inference speed will be AI products' biggest competitive edge in 1–2 years — itsOmSarraf_ · 2026-11-11
- What happened when I moved vector retrieval into Snowflake SQL instead of orchestrating RAG outside — dima806_dima · 2026-11-11
- OpenAI engineer runs Codex for 26 days straight hunting Atlantis, spawning "vibe archaeology" — Polymarket · 2026-11-11
- He replaced a $5,000/month ghostwriter with 11 Claude Skills and earned $47,000 from one post — aryanXmahajan · 2026-11-11
- Refero Styles ships 2,000+ DESIGN.md files to give Claude Code real design taste — bbssppllvv · 2026-11-11