← Back to all tools
Visit site
Overall score
2.4/ 5
SME fit2/5
usage-metered pricing · technical setup
JTBD3/5
moderate JTBD clarity
Integration4/5
API + 7 integrations
Trust1/5
new (<12mo), founded 2025
Quality1/5
no public rating
Compliance2/5
compliance unknown
About
Tinker is a fine-tuning API from Thinking Machines Lab that lets engineers train custom models on top of Llama, Qwen, DeepSeek, GPT-OSS, Moonshot and Nemotron families without managing GPU infrastructure. LoRA-based by default for efficient training. Pay-as-you-go pricing per million tokens trained.
Best for: ML engineers and AI researchers who want to fine-tune open-source LLMs without managing GPU clusters or wiring distributed-training infrastructure themselves.
Pricing
Pay-as-you-go (small models)
- Monthly
- n/a
- Annual /mo
- n/a
- Billing
- usage_based
- Notes
- Llama 3.2 1B fine-tuning at $0.09/M tokens;LoRA training;Checkpoint download API;Self-serve API access · Smallest model tier. Cheap for prototyping fine-tunes.
Pay-as-you-go (large MoE)
- Monthly
- n/a
- Annual /mo
- n/a
- Billing
- usage_based
- Notes
- Qwen3.5-397B-A17B with 256K context at $12/M tokens;Storage $0.10/GB-month;Distributed training · Largest models tier. Limited-time 50% discount available on select NVIDIA models.
| Tier | Monthly | Annual /mo | Billing | Notes |
|---|---|---|---|---|
| Pay-as-you-go (small models) | n/a | n/a | usage_based | Llama 3.2 1B fine-tuning at $0.09/M tokens;LoRA training;Checkpoint download API;Self-serve API access · Smallest model tier. Cheap for prototyping fine-tunes. |
| Pay-as-you-go (large MoE) | n/a | n/a | usage_based | Qwen3.5-397B-A17B with 256K context at $12/M tokens;Storage $0.10/GB-month;Distributed training · Largest models tier. Limited-time 50% discount available on select NVIDIA models. |
Key features
- LoRA-based fine-tuning of open-source models
- Wide model coverage (Llama, Qwen, DeepSeek, GPT-OSS, Moonshot, Nemotron)
- Pay-as-you-go pricing per million tokens
- Distributed training on GPU clusters abstracted away
- Forward and backward pass + optimizer step + sampling primitives
- Model checkpoint downloads via API
Integrations
LlamaQwenDeepSeekGPT-OSSMoonshot KimiNVIDIA NemotronHugging Face
Trust & compliance
- Stage range
- Seed → Growth
- Founded
- 2025
- Status
- active
- SOC 2
- unknown
- GDPR
- unknown
- Data residency
- unknown
- External rating
- n/a
- Last verified
- May 2026
Reviews
Be the first to share your experience.