Can I use Thomson 1.0 Small in a commercial product?

No. Thomson-1.0-Small is published under the PolyForm Strict 1.0.0 license, which Thomson Reuters describes as being for academic and non-commercial use. The company released it so legal and AI academics could evaluate and validate the model directly, and says it will keep opening it to external parties over the coming weeks and months.

How do I download and run Thomson 1.0 Small?

Thomson-1.0-Small lives at huggingface.co/thomsonreuters/Thomson-1.0-Small and loads through standard tooling. The model card lists Hugging Face Transformers, vLLM, SGLang and TokenSpeed as compatible runtimes, and gives a two-line AutoTokenizer plus AutoModelForCausalLM snippet. Weights are BF16 with 3B parameters active per token, so it is far lighter to serve than its 35B total size suggests.

Which Thomson Reuters products actually use the Thomson model?

Thomson lands first in Tabular Analysis inside CoCounsel Legal, the high-volume document review feature used by law firms and corporate legal departments, arriving in an upcoming release. Thomson Reuters plans to extend it across the legal and tax portfolio, and says it will keep a multi-model strategy that still routes some work to OpenAI and Anthropic models.

Is Thomson as good as a general frontier model?

Thomson Reuters claims performance on par with the latest frontier models across a range of tasks, with particular strength in following complex, multi-part professional instructions. The published model card reports a 82.6 average on tax tasks and 75.2 on legal tasks. The company notes it has trained on less than 10% of its proprietary content so far.

Thomson Reuters · 2026-08-24 · major

Thomson 1.0 Small — Thomson Reuters ships its own 35B legal and tax model

Thomson Reuters released Thomson, its first in-house language model, and put a small 35B version on Hugging Face with open weights for academic and non-commercial use. Training cost $40 million.

Hugging Face model card card for thomsonreuters/Thomson-1.0-Small

A legal and tax specialist model, trained on Westlaw-grade content, with a small open-weight version anyone can download.

Key specs

Tax domain average	82.6
Legal domain average	75.2

Quick facts

Maker	Thomson Reuters
Size	35B total, 3B active (MoE)
Context window	262,144 tokens
Base model	tri-fair-lab/Snowdon1.1-Small
License	PolyForm Strict 1.0.0
Training spend	$40 million (talent and compute)
First product	Tabular Analysis in CoCounsel Legal

What is it?

Thomson is Thomson Reuters' first language model built in-house rather than bought from a frontier lab. The company started from an open-source foundation and spent $40 million on talent and compute to specialize it on decades of its own legal, tax, regulatory and news content. A smaller version, Thomson-1.0-Small, is published on Hugging Face with open weights for academic and non-commercial use.

How does it work?

Thomson-1.0-Small is a mixture-of-experts model with 35B total parameters and 3B active per token, derived from tri-fair-lab/Snowdon1.1-Small and shipped in BF16 with a 262,144-token context. Training used roughly 1.63 × 10²³ FLOP over 35,207 B200 GPU-hours, and the model card also describes an agentic Deep Research setup meant to cut hallucinations by forcing tool use and citations.

Why does it matter?

Owning the model means Thomson Reuters controls the inference bill for high-volume document review instead of paying frontier-lab API rates. The open-weight small version also gives legal and AI researchers something they rarely get: a real, downloadable domain model to poke at. Thomson Reuters says it keeps using OpenAI and Anthropic models elsewhere in the platform.

Who is it for?

legal and tax technologists, domain-model researchers

Frequently asked questions

Can I use Thomson 1.0 Small in a commercial product?: No. Thomson-1.0-Small is published under the PolyForm Strict 1.0.0 license, which Thomson Reuters describes as being for academic and non-commercial use. The company released it so legal and AI academics could evaluate and validate the model directly, and says it will keep opening it to external parties over the coming weeks and months.
How do I download and run Thomson 1.0 Small?: Thomson-1.0-Small lives at huggingface.co/thomsonreuters/Thomson-1.0-Small and loads through standard tooling. The model card lists Hugging Face Transformers, vLLM, SGLang and TokenSpeed as compatible runtimes, and gives a two-line AutoTokenizer plus AutoModelForCausalLM snippet. Weights are BF16 with 3B parameters active per token, so it is far lighter to serve than its 35B total size suggests.
Which Thomson Reuters products actually use the Thomson model?: Thomson lands first in Tabular Analysis inside CoCounsel Legal, the high-volume document review feature used by law firms and corporate legal departments, arriving in an upcoming release. Thomson Reuters plans to extend it across the legal and tax portfolio, and says it will keep a multi-model strategy that still routes some work to OpenAI and Anthropic models.
Is Thomson as good as a general frontier model?: Thomson Reuters claims performance on par with the latest frontier models across a range of tasks, with particular strength in following complex, multi-part professional instructions. The published model card reports a 82.6 average on tax tasks and 75.2 on legal tasks. The company notes it has trained on less than 10% of its proprietary content so far.

Try it

AutoModelForCausalLM.from_pretrained("thomsonreuters/Thomson-1.0-Small", dtype="auto")