Thomson
Fiduciary-Grade AI for Professional Work
Context window
262000
Input / 1M tokens
Not sold via public API
Output / 1M tokens
Not sold via public API
Provider
Thomson Reuters
Thomson-1-Large runs inside Thomson Reuters products (launched in CoCounsel Legal Tabular Analysis, August 2026) with no public API. The smaller Thomson-1.0-Small has open weights on Hugging Face under the non-commercial PolyForm Strict 1.0.0 license, with a native 262,144-token context. · Data verified 2026-10-08
Thomson is Thomson Reuters' first proprietary large language model developed in-house on an open-source foundation. Trained on decades of proprietary content from Westlaw, Practical Law, Checkpoint and Reuters with a $40 million investment. Purpose-built specifically for professional legal and tax work with domain-specific expertise.
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Available through CoCounsel Legal for Tabular Analysis. Small open-weight version released on Hugging Face for academic and non-commercial use. API access portal in development.
Strengths
- ✓Domain-specific expertise in legal and tax work
- ✓Strong citation quality comparable to frontier models
- ✓Cost-efficient alternative to frontier models
- ✓Fully owned and controlled by Thomson Reuters
Best for developers who...
When to choose it (and when not to)
Reach for Thomson when...
- →When domain-specific legal/tax expertise is critical
- →For organizations wanting sovereignty over AI training and deployment
- →When cost efficiency matters alongside accuracy
Look elsewhere if...
- ✕For general-purpose tasks outside legal/tax domains
- ✕If maximum reasoning breadth across all domains is required
How to use it
Quickstart
Not announcedNot announcedAPI access portal in development
API model id: Thomson-1.0-Large
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| Legal Domain Performance | Competitive with frontier models | Citation quality tested on Canadian employment law |
| Document Review Accuracy | Not quantified | Deployed for Tabular Analysis in CoCounsel Legal |
| Legal and General Capabilities | Competitive with Claude Opus 4.8 | Instruction following score: 0.914 |
| Legal and general benchmarks | Competitive with Claude Opus 4.8; ahead of GPT-5.5, Claude Sonnet 5, and Gemini 3.1 Pro | Assessed across range of legal and general capabilities |
| Legal and general capabilities benchmarks | Competitive with Claude Opus 4.8; ahead of GPT-5.5, Claude Sonnet 5, and Gemini 3.1 Pro | Early internal evaluations; domain-specific legal reasoning |
| Legal and general capabilities | Competitive with Claude Opus 4.8; ahead of GPT-5.5, Claude Sonnet 5, Gemini 3.1 Pro | Vendor-run benchmarks, not independent verification |
| PRBench Legal Hard | Top performer vs Claude Opus 4.8, GPT-5.5, Gemini 3.1 Pro | Thomson led in legal-specific benchmark |
| Instruction Following (IFEval + FollowBench) | Top performer vs Claude Opus 4.8, GPT-5.5, Gemini 3.1 Pro | Composite benchmark |
| Long Context (Infinity Bench) | Top performer vs Claude Opus 4.8, GPT-5.5, Gemini 3.1 Pro | Composite with internal Thomson Reuters benchmarks |
| Instruction Following | 0.914 | Ahead of Claude Opus 4.8, Gemini 3.1 Pro, GPT-5.5 |
Compare Thomson
Compare Thomson with any other model
Build a comparison →All model comparisons →