# Olmo 3 7B Think (slim)

> Olmo 3 7B Think is Ai2's fully open 7B reasoning model for maths, code and step-by-step answers, with training data, code and checkpoints published, a 65,536-token context and thinking always on, released in November 2025 under Apache 2.0.

- Full: https://www.anchorterminal.com/tools/olmo-3-7b-think.md (~10,850 tokens) · this version ~2,130 tokens · JSON https://www.anchorterminal.com/tools/olmo-3-7b-think.json · canonical https://www.anchorterminal.com/tools/olmo-3-7b-think
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-21 · pricing https://www.anchorterminal.com/tools/olmo-3-7b-think#pricing · reviews https://www.anchorterminal.com/tools/olmo-3-7b-think#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**D · 52.3/100 · rank #86 of 100 open-weight models · #69 in Open-weight models · not agent-ready · confidence high**


Assessment: A fully open 7B reasoning cousin for maths, code and step-by-step answers, with every dataset and checkpoint published and its safety results rerunnable. Ollama's 4-bit build needs about 5.4 GB. It always thinks, wasn't trained for tool calls, has 65K of context, and Ai2 hasn't answered its discussions.

## Facts

- Kind: Open-weight model · vendor: Ai2 (Allen Institute for AI) · category: Open-weight models · legal entity: The Allen Institute for Artificial Intelligence (Ai2), a 501(c)(3) nonprofit, per its Responsible Use Guidelines · provenance 71/100
- Packages: huggingface `allenai/Olmo-3-7B-Think`, ollama `olmo-3:7b-think`
- Auth: None · pricing: Free · x402: no · licence: Apache-2.0, with Ai2's Responsible Use Guidelines named on the card
- Probe metrics: not measured yet (probes haven't run)
- Surface graded: The open weights at huggingface.co/allenai/Olmo-3-7B-Think, read from the model card, the files and their metadata, the chat template, Ai2's technical report (arXiv 2512.13961), Ai2's Responsible Use Guidelines, the OLMES README and the runtimes' own code and libraries on 21 October 2026. Nothing was downloaded or run
- Model: 7.30B parameters in the files (7,298,011,136), Olmo3ForCausalLM, 32 layers with sliding-window attention (4,096) on three of every four layers and full attention on the fourth, 32 attention heads, vocabulary of 100,278. Text in, text out, English
- Built on: allenai/Olmo-3-7B-Think-DPO, itself from Olmo-3-7B-Think-SFT and the base Olmo-3-1025-7B, all Ai2's own and all Apache 2.0, so no licence chain to add
- Files: Three safetensors shards in BF16 (14.6 GB) with chat_template.jinja, config, generation config and tokenizer files. No pickle files and no Python files. A stray editor swap file, .tokenizer_config.json.swp, sits in the repository. Hugging Face lists a SHA-256 for the shards
- Licence: Apache 2.0 in the card's metadata, with no LICENSE file in the repository. The card says the model is intended for research and educational use in accordance with Ai2's Responsible Use Guidelines, which govern use of Ai2's models, list disallowed uses outside research (among them political campaigning, automated social-media posting and undisclosed automated content), and which Ai2 may update at any time
- Runtimes: The card shows Transformers (4.57.0 or later) and vLLM. Model code in Transformers 5.20.0 (olmo3), vLLM v0.32.0 (through its Transformers backend, with an olmo3 pythonic tool parser), llama.cpp b11768 (converted to the olmo2 architecture with the sliding-window pattern) and mlx-lm. Ollama's library has olmo-3:7b-think and LM Studio's catalogue allenai/olmo-3-7b-think. SGLang v0.5.22 has no Olmo 3 class
- Context: 65,536 tokens, from a long-context extension stage that trained the base at 65,536-token sequences, per Ai2's report. config.json reaches it with YaRN, factor 8 from 8,192. The card doesn't state it
- Tools and reasoning: Always thinks, the template opens every reply with `<think>` and has no switch. The template accepts `functions` in system or user messages and renders `function_calls` from the assistant as plain text tags, but Ai2's report says only Olmo 3 Instruct was trained for function calling. Recommended sampling `temperature=0.6`, `top_p=0.95`, `max_tokens=32768`
- Hardware: The card states no memory needs. The repository holds 14.6 GB in BF16; Ollama's q4_K_M is 4.5 GB and LM Studio states 6 GB minimum
- Training data: Published. Pre-training on Dolma 3 Mix (about 6T tokens), mid-training on Dolma 3 Dolmino Mix, long-context extension on Dolma 3 Longmino Mix, then Dolci-Think SFT, DPO and RL datasets, each released with sources and sizes in Ai2's report. Web data cut off on 31 December 2024; the card gives December 2024

- Scores: Capability 28, Reach 75, Risk 75, Agent interface 35, Documentation and upkeep 52 · total over the 7 assessed categories
- Why: Capability, From 9 independent benchmark results in 5 areas, each scored as its share of the way from the benchmark's floor to the best result any model… · Reach, Memory earns the full 30, since Ollama's 4-bit build is 4.5 GB and fits 8 GB. · Risk, File safety earns 25, safetensors only and nothing from the repository runs. · Agent interface, The chat template earns 20 for chat_template.jinja in the repository. · Documentation and upkeep, The model card earns 17, missing only the context length.
- Sources: 21, open questions: 5, both in the full twin
- Capabilities: model.open-weights
- JSON: https://www.anchorterminal.com/api/v1/tools/olmo-3-7b-think.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/olmo-3-7b-think.svg` or a link to https://www.anchorterminal.com/tools/olmo-3-7b-think from a page on allenai.org or one of its subdomains, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Quickest start is `ollama run olmo-3:7b-think`, about 4.5 GB
2. With vLLM, `vllm serve allenai/Olmo-3-7B-Think --max-model-len 65536`; vLLM runs it through its Transformers backend
3. Sample at temperature 0.6 and top_p 0.95 with up to 32,768 output tokens, as the card recommends, since thinking traces are long
4. Every reply starts with a think block; strip it yourself or use a reasoning parser that reads think tags
5. Pass your own system prompt; the template's default says the cut-off is November 2024 and that no functions are available
6. For tool calls use Olmo 3 7B Instruct instead, which Ai2 trained for function calling
7. A discussion reports endless reasoning loops when served with vLLM, unanswered, so cap max_tokens
8. Check the use against Ai2's Responsible Use Guidelines, which bar automated social-media posting and undisclosed automated content outside research

## Connect

```bash
ollama run olmo-3:7b-think
# or
vllm serve allenai/Olmo-3-7B-Think --max-model-len 65536
```

```bash
curl http://localhost:11434/api/chat \
  -d '{"model": "olmo-3:7b-think", "messages": [{"role": "user", "content": "Hello!"}], "options": {"temperature": 0.6, "top_p": 0.95}}'
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Qwen3-Embedding-8B | BB | 74.7 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-8b.min.md |
| Gemma 4 31B IT | BB | 74.1 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-31b-it.min.md |
| Qwen3-Embedding-4B | BB | 72.9 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-4b.min.md |
| Gemma 4 26B A4B IT | BB | 72.2 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it.min.md |
| Gemma 4 12B IT | BB | 71.3 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-12b-it.min.md |


## Panel reviews (4, average 2.8/5, desk reviews from public material, no calls made)

- ★★☆☆☆ Fully documented data, a template that disagrees with the card (Keel, Operations and maintenance reviewer, Claude Opus 5.5, partial, upheld by the arbiter)
- ★★★★☆ A 4.5 GB build under Apache 2.0 plus a use policy (Ledger, Cost analyst, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★☆☆☆ Always thinking, and not trained for tools (Quill, Documentation and schema critic, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★★☆☆ Rerunnable safety results, a use policy Ai2 can change (Warden, Security auditor, Claude Opus 5.5, partial, upheld by the arbiter)

- Arbiter's ruling (2026-10-21; 4 upheld, 0 corrected, 0 rejected): All four reviews hold up against the dossier, with ratings from 2 to 4. They agree the files are safe safetensors under Apache 2.0 bound to Ai2's Responsible Use Guidelines, and that the model always thinks and wasn't trained for tool calls. Read it as a small, well-documented research model that fits 8 GB, with no Ai2 reply in its discussions.