Tinker team benchmarks model finetunability: Inkling 2 Small learns faster for less
ziqiao_ma · x · 2026-11-06
A Thinking Machines member shared Tinker's "model finetunability" benchmark: on an RL fine-tuning task built from public earnings-call data (investment analysis with expert judgment), Inkling 2 Small converges faster and reaches higher held-out rubric reward than GLM 5.3 Flash, DeepSeek V4.1 Flash and Qwen 3.5 397B. Experiments ran 200 steps with Tinker Cookbook's AsyncRL and Stable Reinforce loss; Inkling results averaged over 3 seeds, costs estimated from Tinker per-token pricing with prefix-cache-hit assumptions. The team invites users to try it on Tinker, emphasizing they take post-post-trainability seriously.
More from coding & agent
- Conductor shuts down after Meta acquisition; cherry π ships a one-command migration path — sull · 2026-11-06
- LangChain says most of its internal agents rely on async subagents, now in deep agents — Hacubu · 2026-11-06
- SaaStr runs Replit Agent as its 'CMO': 3 marketing ideas a day — amasad · 2026-11-06
- Visa's Trusted Agent Protocol uses passkeys to verify AI shopping agents at checkout — NVIDIAAI · 2026-11-06
- Cohere's North platform secures AI agents across cloud, private and air-gapped deployments — NVIDIAAI · 2026-11-06
- Choosing browsers by how easily Codex can control them — TheZachMueller · 2026-11-06