Commercial use
Allowed under Apache 2.0. Check any separately published model policy and the requirements of the application domain.
16 verified model passports in the current registry use Apache License 2.0.
Apache 2.0 is a permissive software license with broad rights to use, modify and redistribute and an express patent grant. This page keeps that license fact separate from the broader question of whether an AI system is fully open source.
Apache License 2.0 generally allows use, reproduction, modification, public performance/display, sublicensing and distribution of the licensed work, including in commercial settings.
Redistribution is not obligation-free. The license requires a copy of Apache 2.0, prominent notices for modified files, retention of relevant copyright/patent/trademark/attribution notices, and treatment of a NOTICE file as specified when the distribution includes one.
Apache 2.0 also contains an express patent license from contributors. That is one reason it is often operationally attractive for software and model releases compared with very short permissive licenses.
For AI models, however, a model's weight license is only one layer. Training data, accompanying code, acceptable-use policies, hosted-provider terms and application law can be separate questions. OpenWeightModels therefore labels these exact models “Apache 2.0” without automatically calling every surrounding system fully open source.
These are exact model records verified as Apache 2.0 in the current 32-model registry. Adjacent sizes or future releases should not inherit the classification automatically.
| Model | Parameters | Context | Hardware class | Passport |
|---|---|---|---|---|
| Devstral Small 2505Mistral AI | 24B | 128K tokens | 17–32B · high-memory workstation | License record → |
| gpt-oss-120bOpenAI | 116.8B | 131,072 tokens | >80B · datacenter / multi-GPU | License record → |
| gpt-oss-20bOpenAI | 20.9B | 131,072 tokens | 17–32B · high-memory workstation | License record → |
| Granite 3.3 8B InstructIBM | 8B | 128K tokens | ≤8B · consumer/local | License record → |
| Magistral Small 2506Mistral AI | 24B | 128K nominal · 40K recommended for quality | 17–32B · high-memory workstation | License record → |
| Mathstral 7B v0.1Mistral AI | 7B | 32,768 tokens | ≤8B · consumer/local | License record → |
| Mistral Nemo Instruct 2407Mistral AI / NVIDIA | 12B | 128K tokens | 9–16B · workstation/local | License record → |
| Mistral Small 3.1 24B InstructMistral AI | 24B | 128K tokens | 17–32B · high-memory workstation | License record → |
| OLMo 2 13B InstructAi2 | 13B | 4,096 tokens | 9–16B · workstation/local | License record → |
| OLMo 2 32B InstructAi2 | 32B | 4,096 tokens | 17–32B · high-memory workstation | License record → |
| Qwen2.5-Coder-32B-InstructQwen / Alibaba | ~32.5B | 32K native · long-context extension documented by Qwen | 17–32B · high-memory workstation | License record → |
| Qwen3-235B-A22BQwen / Alibaba | 235B | 32,768 native · 131,072 with YaRN | >80B · datacenter / multi-GPU | License record → |
| Qwen3-30B-A3BQwen / Alibaba | 30.5B | 32,768 native · 131,072 with YaRN | 17–32B · high-memory workstation | License record → |
| Qwen3-32BQwen / Alibaba | 32.8B | 32,768 native · 131,072 with YaRN | 17–32B · high-memory workstation | License record → |
| Qwen3-Coder-480B-A35B-InstructQwen / Alibaba | 480B | 256K tokens | >80B · datacenter / multi-GPU | License record → |
| SmolLM3 3BHugging Face | 3B | 64K native · up to 128K extended | ≤8B · consumer/local | License record → |
Allowed under Apache 2.0. Check any separately published model policy and the requirements of the application domain.
Allowed. Modified files must carry prominent notices stating that changes were made when redistributed.
Allowed subject to the license's copy, notice and attribution-preservation requirements. A supplied NOTICE file can create additional notice-handling obligations.
Apache License 2.0 permits commercial use, modification and redistribution subject to its conditions. Separate model policies or applicable law can still matter and should be checked for the exact release.
Not automatically. A weights license does not by itself establish that training data, training code and every component satisfy an Open Source AI definition.
Redistributions must include a copy of the license, preserve required notices, identify modified files, and handle a NOTICE file as specified by the license when one is supplied.
Licensing can differ between sizes, versions or adjacent model families. The exact artifact and license file are the unit of verification.