Billed by usage per million tokens, with prices per model and context length. Payment details are set up under Billing in the Tinker Console.
Billed per million tokens for prefill (input), sample (output) and training
80% discount on cached prefill tokens
Checkpoint storage at USD 0.10 per GB per month
29 model variants including Qwen, Kimi, Nemotron, GLM, DeepSeek, GPT-OSS and Inkling
LoRA fine-tuning with supervised learning and reinforcement learning
Download of saved checkpoints
GPT-OSS-20B: USD 0.18 prefill, USD 0.45 sample and USD 0.396 training per 1M tokens
Qwen3-8B: USD 0.195 prefill, USD 0.60 sample and USD 0.44 training per 1M tokens
Qwen3.5-4B: USD 0.33 prefill, USD 1.005 sample and USD 0.737 training per 1M tokens
GPT-OSS-120B: USD 0.33 prefill, USD 0.84 sample and USD 0.737 training per 1M tokens
Qwen3.6-35B-A3B: USD 0.54 prefill, USD 1.335 sample and USD 1.177 training per 1M tokens
DeepSeek-V3.1: USD 1.695 prefill, USD 4.215 sample and USD 3.718 training per 1M tokens
Kimi-K2.6: USD 2.205 prefill, USD 5.49 sample and USD 4.84 training per 1M tokens
Qwen3.5-397B-A17B: USD 3.00 prefill, USD 7.50 sample and USD 6.60 training per 1M tokens
GLM-5.3 (256K): USD 4.86 prefill, USD 12.15 sample and USD 14.58 training per 1M tokens
Serverless inference (beta) for Inkling-Small and Inkling: from USD 0.30 input and USD 1.20 output per 1M tokens
