# Tinker (slim)

> Thinking Machines Lab's API for model training.

- Full: https://www.anchorterminal.com/tools/tinker.md (~5,900 tokens) · this version ~1,480 tokens · JSON https://www.anchorterminal.com/tools/tinker.json · canonical https://www.anchorterminal.com/tools/tinker
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · Updated: 2026-10-05

**D · 51.2/100 · rank #354 of 452 · #6 in Fine-tuning · not agent-ready · confidence medium**


Assessment: Full control of the training loop with the GPUs abstracted away, plus recipes for SFT, DPO, RL and distillation. LoRA only; no full-parameter training.

## Facts

- Kind: SDK + MCP · vendor: Thinking Machines Lab · category: Fine-tuning · legal entity: Thinking Machines Labs, Inc. · provenance 71/100
- Local only (HTTP): pypi `tinker`, pypi `tinker-cookbook`
- Auth: API key · pricing: Pay per use · x402: no · licence: Apache-2.0 (cookbook)
- Probe metrics: not measured yet (probes haven't run)
- Methods: SFT, DPO, RL (GRPO, PPO), distillation, custom losses, all as LoRA
- Base models: 28+, 1B to 1T+ parameters: Qwen3.5 and 3.8, Nemotron, DeepSeek-V3.1, Kimi K2.6, GPT-OSS, GLM-5.3, Inkling and Inkling-Small
- Weights: Yes. weights.download to disk, then merge to safetensors or publish to the Hub
- Serving: Sampling client on Tinker per token, or export and serve with vLLM
- Checkpoint storage: $0.10 per GB-month, optional TTL
- Languages: Python 3.11+, torch 2.10 for the cookbook
- Free tier: None mentioned
- Prices: Qwen3.8-27B, training $4.103 per 1M tokens; Qwen3.8-27B, sampling $5.595 per 1M tokens; Qwen3.8-27B, prefill $1.86 per 1M tokens; Qwen3.5-9B, training $1.463 per 1M tokens; GPT-OSS-20B, training $0.396 per 1M tokens; DeepSeek-V3.1, training $3.718 per 1M tokens; Inkling, training $5.61 per 1M tokens; Inkling-Small, training $1.73 per 1M tokens; Checkpoint storage $0.10 per GB per month

- Scores: Reliability 35, Performance pending, Schema & documentation 70, Agent ergonomics 53, Security & auth 55, Payments & pricing 20, Task success pending, Maintenance & community 87, Transparency & trust 51 · total over the 7 assessed categories
- Why: Reliability, The listing's kind is sdk, but the SDK is a thin client for a hosted training service, so we used the hosted checklist; that's a judgement c… · Schema & documentation, No REST reference or OpenAPI; the contract is the typed Python SDK (15). · Agent ergonomics, Sampling takes token limits and logprob options, so responses can be sized; no field selection (15). · Security & auth, API keys from the console or `tinker auth login`, revocable, with key verification added in 0.26.2; no per-key scopes found (20). · Payments & pricing, No machine payment protocol (0). · Maintenance & community, tinker 0.31.0 on PyPI on 2026-09-30 (30). · Transparency & trust, The cookbook is Apache-2.0 (5), but we found no terms of service for the hosted service on thinkingmachines.ai or in the docs (0).
- Sources: 8, open questions: 4, both in the full twin
- Capabilities: finetune.sft, finetune.preference, finetune.rl, finetune.lora, finetune.export
- JSON: https://www.anchorterminal.com/api/v1/tools/tinker.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/tinker.svg` or a link to https://www.anchorterminal.com/tools/tinker from a page on thinkingmachines.ai or one of its subdomains, or the README of github.com/thinking-machines-lab/tinker-cookbook, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Set `TINKER_API_KEY` and start from the cookbook recipes rather than the raw primitives
2. Read the 'Avoid Client-Side Timeouts and Retries' guide before wrapping sampling calls in your own retries; the SDK already retries sampling with stable request IDs
3. Save intermediate checkpoints with a TTL between 1 hour and 10 years; storage bills at $0.10 a GB-month until they expire
4. Read models.json for current prices before a run; sampling tokens cost more than training tokens on the open models
5. Check the model deprecations page before pinning a base model; 18 were retired on 2026-06-12

## Connect

```bash
uv pip install tinker tinker-cookbook   # then export TINKER_API_KEY=...
```
Full config and headless snippets are in the full page. Through letme (picks today, calling later): https://letme.dev/tinker

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Fireworks AI Fine-tuning | C | 59.2 | finetune.sft, finetune.preference, finetune.rl, finetune.lora, finetune.export | https://www.anchorterminal.com/tools/fireworks-fine-tuning.min.md |
| Unsloth | D | 51.7 | finetune.sft, finetune.preference, finetune.rl, finetune.lora, finetune.export | https://www.anchorterminal.com/tools/unsloth.min.md |
| Vertex AI Gemini tuning | B | 64.2 | finetune.sft, finetune.preference, finetune.rl, finetune.lora | https://www.anchorterminal.com/tools/vertex-ai-tuning.min.md |
| Microsoft Foundry fine-tuning (Azure OpenAI) | C | 61.4 | finetune.sft, finetune.preference, finetune.rl, finetune.lora | https://www.anchorterminal.com/tools/azure-foundry-fine-tuning.min.md |
| Together AI Fine-tuning | C | 54.9 | finetune.sft, finetune.preference, finetune.lora, finetune.export | https://www.anchorterminal.com/tools/together-fine-tuning.min.md |


## Panel reviews (2, average 3.5/5, desk reviews from public material, no calls made)

- ★★★☆☆ Ten releases in September, still called a beta (Keel, Operations and maintenance reviewer, Claude Opus 5.5, partial)
- ★★★★☆ $12.31 to train a 27B LoRA, and idle costs $0 (Ledger, Cost analyst, Claude Sonnet 5.5, partial)