Our take: Tinker is for ML researchers and teams that want to write their own training data and algorithm code for LoRA fine-tuning of open models while Thinking Machines runs the GPU infrastructure.
Main limitation: It is aimed at people writing training code, billing is usage-based per million tokens with checkpoint storage at $0.10/GB-month, and some models retire on set dates.
By the AI Tool Atlas editors.
Based on the vendor's official documentation; we have not tested this product hands-on. What “documented overview” means
Is Tinker (Thinking Machines Lab) right for you?
Tinker is for ML researchers and teams that want to write their own training data and algorithm code for LoRA fine-tuning of open models while Thinking Machines runs the GPU infrastructure. Based on the vendor's documentation, not on a hands-on test.
Best for: ML researchers
Also a good fit for
Teams fine-tuning open models
Consider another tool if requires sign-up and account
Or if
Aimed at researchers writing training code
Checkpoint storage billed at $0.10/GB-month
Tinker (Thinking Machines Lab) pricing and usage limits
Prices checked Sep 29, 2026 on the official US pricing page. Official pricing ↗ Prices and limits change often.
For: Researchers and developers fine-tuning open models
Limits
Checkpoint storage $0.10 per GB-month
Per-model token rates are in the docs, not captured
Includes
forward_backward, optim_step, sample, save_state API
LoRA fine-tuning of open models
1 more
Downloadable checkpoints
API included
Usage caps, credits and extra costs
Checkpoint storage billed at $0.10/GB-month
Not stated
What is Tinker (Thinking Machines Lab), and how does it work?
Tinker is a training API for researchers and developers. You write the data and algorithm code and Tinker handles scheduling, resource management and GPU infrastructure.
It exposes functions such as forward_backward, optim_step, sample and save_state, and supports a range of open-weight models using LoRA.
Tinker (Thinking Machines Lab)'s official site: how the product presents itself. thinkingmachines.ai, captured Oct 2, 2026
What you provide and what you get back
What you give Tinker (Thinking Machines Lab) and what comes back, from its documentation
You provide
Supervised learning datasets or RL environments
Training algorithm code
A chosen open-weight base model
You get back
LoRA adapters and saved checkpoints
Sampled tokens for evaluation or RL
Downloadable model weights
Where you can use it
Platforms
API
Account
Required
Method
LoRA
Core functions
forward_backward, optim_step, sample, save_state
Models
Qwen, Kimi, DeepSeek, GPT-OSS, Nemotron, GLM
Storage price
$0.10 per GB-month
Export
Checkpoint download
Best use cases for Tinker (Thinking Machines Lab)
Scenarios are worked out by our editors from the documentation; the actual result depends on your input and plan.
Fine-tuning open models on production traces
Who
ML engineers
You provide
Anonymized traces such as preferred and rejected tool calls
Expected result
A smaller specialized model (e.g. a search subagent)
A training API from Thinking Machines Lab for fine-tuning open models with LoRA while it handles scheduling and GPU infrastructure. source (opens in a new tab)
How is Tinker priced?
Usage-based in USD per million tokens, with checkpoint storage at $0.10 per GB-month. source (opens in a new tab)
Is my data used to train Thinking Machines models?
Not verified: Per-million-token rates; Free credits; Privacy retention; US availability.
How we documented and maintain this page
Every fact about Tinker (Thinking Machines Lab) above comes from the 2 official pages listed here, last compared on Sep 29, 2026; we have not run Tinker (Thinking Machines Lab) ourselves, so it has no editorial score. How we check and re-check pages.
Article updated Oct 10, 2026; prices checked Sep 29, 2026. Outbound links to vendors are plain links; we do not earn commissions. How we review · Report an error
