Approved models
6
Records published from this collection
Provider information
lab
Review approach
Verify each model’s source, terms, capabilities, and current availability
Ai2 publishes the Olmo model family alongside research artifacts for studying how language models are trained and evaluated. A downloadable checkpoint and supporting training material serve different purposes.
Olmo 3.1 32B Instruct is a starting point for ordinary assistant tasks. Olmo 3.1 32B Think has a reasoning-oriented interface. The 7B Instruct and 7B Think releases offer smaller checkpoints for a separate deployment evaluation.
Use the supplied chat template and keep the output budget visible. Reasoning output, final-answer parsing and runtime settings can change latency and acceptance behavior. A smaller download can still miss the quality target for a particular task.
Training artifacts help explain provenance and reproduce research choices. They do not establish that a model will meet an application’s accuracy requirements. For a document assistant, include missing answers and conflicting versions. For coding, require an executable result.
These listed checkpoints use Apache-2.0 licensing. Review their exact repository configuration before quantization or serving. Record local compute costs or verify a third-party host independently; no managed price is inferred from the availability of weights.
Review each record individually; collection membership does not imply a recommendation or guarantee of availability.
| allenai/MolmoPoint-Vid-4B Ai2 (Allen Institute for AI) | 5.0 | — | — | — | — |
| allenai/specter2_base Ai2 (Allen Institute for AI) | 5.0 | — | — | — | — |
| Olmo 3 7B Instruct Ai2 (Allen Institute for AI) | 0.0 | — | — | — | — |
| Olmo 3 7B Think Ai2 (Allen Institute for AI) | 15.0 | — | — | — | — |
| Olmo 3.1 32B Instruct Ai2 (Allen Institute for AI) | 0.0 | — | — | — | — |
| Olmo 3.1 32B Think Ai2 (Allen Institute for AI) | 15.0 | — | — | — | — |
1–6 of 6 models · select a column to sort
About this order: Recommended shows our editorial shortlist first, followed by recent releases. The spec score is a 0–100 figure SyncDev calculates from each vendor's published specifications — context window, reasoning support, input modalities, tool calling, maximum output and how recently the model shipped. It measures breadth of capability, not benchmark performance, so a higher-scoring model is not automatically the better choice for your task.