Thomson-1-Large Launch — Legal AI Pricing Impact
Thomson Reuters says its new legal AI model rivals frontier systems. See its benchmark results, August rollout, and still-unpublished pricing.
By AI Pricing Guru Editorial Team
AI Pricing Guru articles are maintained by the editorial workflow behind the site: daily pricing snapshots, provider source checks, and review passes for model launches, subscription limits, and billing changes.
TL;DR
- Thomson Reuters unveiled Thomson-1-Large, a proprietary model trained for legal and professional work.
- Its first deployment is scheduled for August as the default model in CoCounsel Legal's Tabular Analysis feature.
- Thomson Reuters has not announced a standalone API, token rate, or separate model surcharge.
- The vendor's benchmark is promising, but it mixes reasoning settings and includes internal long-context tests; buyers should wait for task-level cost and accuracy data.
Public alternatives: cost for 1M input and 200K output tokens
USD per 1M tokens. Input and output rates are charted separately.
Public API prices for the benchmarked alternatives
| Model | Provider | Input / 1M | Cached / 1M | Output / 1M |
|---|---|---|---|---|
| Claude Opus 4.8 | anthropic | $5.00 | $0.5 | $25.00 |
| GPT-5.5 | openai | $5.00 | $0.5 | $30.00 |
| Gemini 3.1 Pro | $2.00 | $0.2 | $12.00 |
Built from pricing.json at publish time.
Thomson Reuters announced early results for Thomson-1-Large on July 31, positioning the proprietary model as a smaller, domain-specialized alternative to frontier systems for legal and tax work. The announcement was detected in today’s model-monitoring cycle.
The company says Thomson-1-Large performs competitively with Claude Opus 4.8 and ahead of GPT-5.5, Claude Sonnet 5, and Gemini 3.1 Pro across its selected evaluation set. That is a benchmark claim, not yet a price-performance result: Thomson Reuters published no standalone API rate or inference-cost figure.
The live chart and table above show current public token prices for three models included in Thomson Reuters’ comparison. They exclude Thomson-1-Large because no public token rate exists.
What launched
Thomson-1-Large is the first model in a new Thomson Reuters model family. It follows the company’s 2024 acquisition of Safe Sign Technologies and combines model development with proprietary professional content and expert-written evaluations.
The first production deployment is due in August inside Tabular Analysis in CoCounsel Legal. Thomson-1-Large will become that feature’s default model for structured, high-volume document review. Thomson Reuters says it plans broader legal and tax integrations over the following year.
This is not a general self-serve model launch. The announcement gives no public model endpoint, context-window specification, rate limits, regional availability, or downloadable weights.
Benchmark results
Thomson Reuters reported these head-to-head scores. Higher is better within each row.
| Evaluation | Thomson-1-Large | Gemini 3.1 Pro | Claude Opus 4.8 | GPT-5.5 |
|---|---|---|---|---|
| Stanford LegalBench | 0.823 | 0.843 | 0.818 | 0.832 |
| PRBench Legal Hard | 0.352 | 0.293 | 0.315 | 0.333 |
| Harvey Legal Agent Benchmark | 0.857 | 0.555 | 0.869 | 0.781 |
| Instruction following composite | 0.914 | 0.848 | 0.861 | 0.885 |
| Reasoning composite | 0.684 | 0.748 | 0.737 | 0.589 |
| Coding | 0.399 | 0.500 | 0.598 | 0.414 |
| Long-context composite | 0.753 | 0.750 | 0.741 | 0.703 |
The table supports a narrower conclusion than the headline: Thomson-1-Large led three of seven reported rows and came close to the best model on two legal evaluations. It did not lead Stanford LegalBench, Harvey’s legal-agent benchmark, reasoning, or coding.
Pricing impact
The immediate pricing signal is vertical integration, not a confirmed token discount. Thomson Reuters says its model is a fraction of the size and operating cost of many general-purpose systems, but it did not disclose parameters, serving cost, customer price, or whether CoCounsel plans will change.
For buyers, the relevant comparison is the total cost of a completed legal workflow. A domain model can justify a premium if it reduces missed clauses, review time, escalation, or repeat runs. A lower internal inference bill does not guarantee a lower customer price, especially when the model is bundled into a professional product rather than sold as an API.
Until commercial terms appear, treat CoCounsel access as the purchase route and the public models above as build-your-own alternatives. Use our token calculator to model those alternatives, then add retrieval, document parsing, human review, and compliance costs.
Benchmark caveats
This is a vendor-run evaluation, not an independent audit. Thomson-1-Large used test-time scaling; Gemini 3.1 Pro and Claude Opus 4.8 used reasoning mode; GPT-5.5 used non-reasoning mode. Those settings can change both quality and inference cost.
The long-context score also combines a public benchmark with Thomson Reuters’ internal tests. The reasoning and instruction-following rows are composites, which makes the headline useful for screening but insufficient for procurement.
The strongest next evidence would be blind evaluation on buyer documents, latency, human-review time, and cost per correctly completed matter. For a broader framework, see our AI API pricing comparison and API cost-calculation guide.
What legal AI buyers should do
- Ask whether Thomson-1-Large changes CoCounsel pricing, quotas, retention, data residency, or contractual terms.
- Request results on your document types and define an error-cost threshold before migration.
- Compare cost per accepted review, not model size or a blended benchmark score.
- Keep a tested fallback for work outside Thomson-1-Large’s domain and for service interruptions.
- Re-run the evaluation when Thomson Reuters expands training beyond the currently disclosed fraction of its content.
Developers building an independent stack can compare live Anthropic pricing, OpenAI pricing, and Google AI pricing before selecting a general model plus retrieval layer.
Bottom line
Thomson-1-Large is a credible new legal-model entrant with a concrete August deployment, but not yet a publicly priced API product. Its reported legal and instruction-following results justify a buyer evaluation; they do not establish a cost advantage.
The decision point is straightforward: CoCounsel customers should ask for task-level economics, while API teams should keep budgeting against published alternatives until Thomson Reuters exposes commercial terms.
Source: Thomson Reuters’ official announcement, “Thomson Reuters Built Its Own AI Model That Now Ranks Among the World’s Best”, published July 31, 2026.