TinkerToy Computer invented by Daniel Hillis and Brian Silverman

Tinker is a training API for researchers

Control every aspect of model training and fine-tuning while we handle the infrastructure.

Sign up Sign in Docs

Your ideas in four functions

forward_backward

Perform a forward pass and a backward pass, accumulating the gradient.

optim_step

Update weights based on the accumulated gradient.

sample

Generate tokens for interaction, evaluation, or RL actions.

save_state

Save training progress for resumption.

Supported models

QWEN
Qwen3.5-4B Dense Qwen3.5-27B Dense Qwen3.5-35B-A3B MoE Qwen3.5-397B-A17B MoE Qwen3-4B-Instruct-2507 Dense Qwen3-8B-Base Dense Qwen3-8B Dense Qwen3-30B-A3B-Base MoE Qwen3-30B-A3B MoE Qwen3-30B-A3B-Instruct-2507 MoE Qwen3-VL-30B-A3B-Instruct MoE Qwen3-32B Dense Qwen3-235B-A22B-Instruct-2507 MoE Qwen3-VL-235B-A22B-Instruct MoE

GPT-OSS
GPT-OSS-120B MoE GPT-OSS-20B MoE

LLAMA
Llama-3.2-1B Dense Llama-3.2-3B Dense Llama-3.1-8B Dense Llama-3.1-8B-Instruct Dense Llama-3.1-70B Dense Llama-3.3-70B-Instruct Dense

DEEPSEEK
DeepSeek-V3.1 MoE DeepSeek-V3.1-Base MoE

MOONSHOT
Kimi-K2-Thinking MoE Kimi-K2.5 MoE

NVIDIA
Nemotron-3-Nano-30B-A3B-BF16 MoE Nemotron-3-Super-120B-A12B-BF16 MoE
LoRA update visualization

Tinker uses LoRA

LoRA fine-tunes models by training a small add-on instead of changing all the original weights.

Read blog post

Tinker cookbook

Explore realistic examples and common abstractions for fine-tuning language models with the Tinker API.

View cookbook

Tinker lets researchers focus on datasets, algorithms, and environments without the complexities of compute and infrastructure.

Tyler Griggs

Tinker lets us focus on the research, rather than spending time on engineering overhead. That's something no amount of raw GPU credits can substitute.

Ziran Yang, Yong Lin, Chi Jin

The training infrastructure has been abstracted away, which makes focusing on our data and evals far easier. Tinker has made it easy to jump into RL work.

Jason Liu

Tinker has been reliable for quickly iterating without worrying about hardware or infrastructure.

Eric Gan

FAQs

Join here. If you're a university or organization looking for wide scale access, contact [email protected].
Tinker is a flexible API for efficiently fine-tuning open source models with LoRA. It's designed for researchers and developers who want flexibility and full control of their data and algorithms without worrying about infrastructure management.
LoRA is an efficient approach to fine-tuning that trains a streamlined adapter instead of updating all base model weights. Our research demonstrates that with the right setup, LoRA matches the learning performance of full fine-tuning while providing more flexibility and requiring less compute.
Tinker handles scheduling, tuning, resource management, and infrastructure reliability so you can focus on the training data and algorithms. Behind the scenes, Tinker orchestrates distributed training on powerful GPU clusters for efficient utilization.
A dataset of supervised learning examples or reinforcement learning environments. After picking a base model to train on, the Tinker API provides simple functions to compute gradients, update the weights, and sample outputs from the trained model. See our cookbook for examples to get started.
Tinker is currently available for a broad selection of open-source models, ranging from compact models like Llama-3.2-1B to large MoEs like Qwen3.5-397B-A17B. We plan to expand our model lineup with even more choices soon.
Your data is used solely to fine-tune your models. We do not use your data to train our own models.
Yes, we have an API endpoint that lets you download any checkpoint you've saved.
Follow Tinker on X for API updates, new model and feature announcements, and weekly roundups of featured projects from the Tinker community.

Tinker uses a pricing plan that reflects compute usage. All prices are in USD per million tokens. Storage is charged at $0.10/GB-month.

Model Context Length Prefill Sample Train
Nemotron-3-Nano-30B-A3B-BF16Limited-time 50% discount 64K $0.13 $0.33 $0.40
Nemotron-3-Super-120B-A12B-BF16Limited-time 50% discount 64K $0.38 $0.96 $1.16
256K $0.76 $1.92 $2.32
Qwen3.5-397B-A17B 64K $2.00 $5.00 $6.00
256K $4.00 $10.00 $12.00
Qwen3.5-35B-A3B 64K $0.36 $0.89 $1.07
Qwen3.5-27B 64K $1.24 $3.73 $3.73
Qwen3.5-4B 64K $0.22 $0.67 $0.67
Qwen3-4B-Instruct-2507 32K $0.07 $0.22 $0.22
Qwen3-8B 32K $0.13 $0.40 $0.40
Qwen3-30B-A3B 32K $0.12 $0.30 $0.36
Qwen3-VL-30B-A3B-Instruct 32K $0.18 $0.44 $0.53
Qwen3-32B 32K $0.49 $1.47 $1.47
Qwen3-235B-Instruct-2507 32K $0.68 $1.70 $2.04
Qwen3-VL-235B-A22B-Instruct 32K $1.02 $2.56 $3.07
Llama-3.2-1B 32K $0.03 $0.09 $0.09
Llama-3.2-3B 32K $0.06 $0.18 $0.18
Llama-3.1-8B 32K $0.13 $0.40 $0.40
Llama-3.1-70B 32K $1.05 $3.16 $3.16
DeepSeek-V3.1 32K $1.13 $2.81 $3.38
GPT-OSS-120B 32K $0.18 $0.44 $0.52
128K $0.63 $1.54 $1.82
GPT-OSS-20B 32K $0.12 $0.30 $0.36
Kimi-K2-Thinking 32K $0.98 $2.44 $2.93
Kimi K2.5 32K $1.47 $3.66 $4.40
128K $5.15 $12.81 $15.40