AI BriefingThinking Machines Lab Releases Active 17B Inkling 2 Small Tinker and Plans to Open Weights
·2026.11.07 09:30
Key point
Inkling 2 Small achieved the final score of the first generation in just 75 steps of RL fine-tuning.
Details
- Announced the second-generation model Inkling 2 Small, featuring a 594B total, 17B active MoE architecture and supporting text and image inputs
- Inference and fine-tuning provided on the Tinker platform, with Apache 2.0 weights scheduled for release next week after external safety testing
- Reached the first-generation final score in 75 steps under the same RL configuration
- Scored 30.6 on AA Index v5.0 and 70.5% on JobBench (ranked 1st), demonstrating strengths in practical work, while scoring 82.2% on SWEBench Verified, indicating a weakness in coding agents
- Application of a new tokenizer increased the number of Korean tokens by approximately 31%, leading to higher training token and serverless inference costs
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Thinking Machines AI introduces Inkling 2 Small, planning Apache 2.0 open weights release next week