Thinking Machines' Tinker turns 1: fine-tuned Qwen3 beats GPT-5.5 at 1/14th the cost
thinkymachines · x · 2026-10-16
Thinking Machines Lab marks one year of Tinker, its platform for fine-tuning frontier-grade models customers own, with a refreshed website and new quickstart flow.
Case studies:
- Bridgewater fine-tuned Qwen3-235B for financial information filtering: 84.66% accuracy across six held-out tasks vs GPT-5.5's 78.2% (30% fewer errors) at roughly 1/14th the inference cost.
- Lightning Rod Labs fine-tuned gpt-oss-120b to predict clinical events (dialysis, transfusion, ventilation) from raw notes, cutting prediction error 27% and calibration error 69%, matching GPT-5.
Targeted use cases span classification, forecasting, search agents, continual learning, and code generation.
Related event: Thinking Machines Marks Tinker's First Year of Accessible Fine-Tuning(3 posts)→
More from Companies & People
- Dario Amodei says AI better than humans at almost everything may be 2-3 years away — PolarBearby · 2026-10-16
- Jensen Huang: Musk's AI buildout moved from K's to M's, now a CSP-tier NVIDIA customer — elonmusk · 2026-10-16
- Vercel CEO Rauch and Shopify CEO Tobi Lütke share stage at Ship SF — evilrabbit_ · 2026-10-16
- Australia's AERO guidance: avoid AI entirely in first three years of primary school — lmoroney · 2026-10-16
- Alexandr Wang: Meta Superintelligence Labs is in South Bay, and strangers talking to you means something's up — HanchungLee · 2026-10-16
- Accenture caps Claude reimbursement at $29/month while employees spend up to $1,000 — inductionheads · 2026-10-16