Thomson Reuters released Thomson, a family of models built by retraining open-weight Qwen models. The company calls its method continual learning, a staged retraining that adds skills without losing old ones. A team of under three dozen staff did the work on a small cluster in three months. Thomson-1.0-Large scores near the strongest proprietary models tested on an internal cross-domain suite. It leads on instruction following, general agent tasks and political neutrality. It trails on coding, maths and multilingual tasks. Thomson-1.0-Small, at 35 billion parameters, is released with open weights. Blind expert reviewers preferred Thomson's answers over rival systems in most paired comparisons. Thomson Reuters ran the evaluations itself and notes it cannot see hidden safeguards inside rival hosted models. Partners include Imperial College London, DatologyAI and Lambda.
What changed
Frontier-level model building was assumed to need very large compute and staff budgets.
What it unlocks
Institutions can upgrade an open-weight model to near-frontier scores on legal, tax and research work with a small team.
- under USD 450,000 final training run
- ~USD 40M total development cost
- 368 B200 GPUs at peak
- 3 months from first experiments
What you need to act on it
- access to a few hundred B200-class GPUs
- a team of around three dozen engineers and scientists
- permissively licensed training data
Sources