• fine-tuning — Taking a trained model and training it a bit more on your own examples so it gets better at one specific job.

Tinker's fine-tuning API is now open to everyone, adds Kimi K2 Thinking and Qwen3-VL vision input, and exposes an OpenAI compatible sampling endpoint so in-progress checkpoints drop straight into existing clients.

• postThinking Machines opens a research blog, starting with inference nondeterminism
• postThinking Machines argues for staged access, using Inkling as the test case
• postOn-policy distillation: RL's error correction at SFT prices
• postExpert judgment in the reward function beats scaffolds on text-to-SQL

• articleTinker General AvailabilityThinking Machines
• postKimi K2.5 arrives on Tinker, text input firstThinking Machines
• postTinker adds gpt-oss and DeepSeek-V3.1 models to its fine-tuning lineupThinking Machines

Checking sign-in…

Loading comments…