← Back to the wire

Thomson Reuters bets $40M on owning its AI instead of renting from OpenAI or Anthropic

AnnouncementModelAug 24, 2026

Thomson Reuters spent about $40 million over two years building "Thomson," its first in-house legal language model, on Alibaba's Qwen. In testing, Thomson only outperforms GPT-5.4 when given access to exclusive company content; with web access alone it trails 0.53 to 0.65. A safety-focused intermediate version, "Snowdon," was developed with Imperial College. Thomson debuts in CoCounsel document review, and a smaller open-weight version will be released under a non-commercial license.

Receipt № 15451 source · awaiting confirmation ◐

Evidence

1source· awaiting independent confirmation

No score is assigned. Sources and their independence are shown in the citation chain below.

Citation chain · 1 source

OpenAICompanyAlibabaCompanyAnthropicCompanyQwenModelGPT-5.4ModelThomson ReutersCompanyImperial CollegeCompanyThomsonModelQwen3.5-397BModelSnowdonModel
Canonical: https://the-decoder.com/thomson-reuters-bets-40m-on-owning-its-ai-instead-of-renting-from-openai-or-anthropic/