AllocateOpen models you can fine-tune commercially
26 open-weight language models on the catalog carry MIT or Apache 2.0 licenses, which permit commercial fine-tuning with no conditions on the resulting weights. They run from OpenAI GPT-OSS 20B at $0.05 per million input tokens up to 397B-parameter MoE bases.
Provider list prices from the Allocate catalog, checked 2026-07-08. Billed price on Allocate is list plus the 7% transaction fee.
The commercial fine-tuning test
Two questions decide whether a base is safe to build a business on: can you use the model commercially, and do you own what training produces. MIT and Apache 2.0 answer yes to both without qualification; community licenses usually answer yes with conditions; unlisted licenses answer nothing, which is its own answer.
This list applies that test mechanically: open weights, permissive license, ranked by list price. On Allocate, whatever you train on these bases stays inside your isolation boundary and leaves with you if you go.
Common questions
Which open models can I fine-tune for commercial use?
Any model under MIT or Apache 2.0: OpenAI GPT-OSS 20B, DeepSeek R1 Distill Qwen 1.5B, Ministral 3 14B Instruct 2512, and the rest of this list. Community licenses like Llama’s also permit commercial fine-tuning for most companies, with conditions worth reading once.
Do I owe anything on the fine-tuned weights?
Under MIT and Apache 2.0, no: attribution on the base model is the only obligation, and the trained weights are yours. On Allocate they are also contractually yours and exportable at any time.
Related
Every model here sits behind one key on Allocate: route by name, meter per route, and swap the model in one click.