Meta’s Muse Wonder Beats OpenAI On Real Lawyer Work In New Test
Meta‘s Muse Wonder just took the top overall quality score in a monthly test of how AI handles real legal work, beating rival models from OpenAI in a market where law firms are deciding which AI to trust with client cases.
Key Takeaways
- Muse Wonder ranked highest on overall quality in Legora’s November results for legal work
- It led four task types: analysis and answering, extraction, drafting, and editing
- OpenAI’s GPT-6.2 Astra led comparison and review tasks at a cost per case 20% below average
- OpenAI’s GPT-6.2 Sol delivered above-average quality at 78% below the average cost per case
Meta posted on X that legal work is “a tough test for AI” and called the result “a strong showing on the kinds of complex tasks businesses need help with.”
The ranking comes from Legora, a legal AI platform used by law firms. Each month it publishes results from its BAR benchmark, which compares how AI models perform on real-world legal tasks inside its product.
Where Meta Won And Where OpenAI Won
| Model | Maker | Where It Led | Cost Per Case |
|---|---|---|---|
| Muse Wonder | Meta | Highest overall quality, plus analysis and answering, extraction, drafting and editing | Not disclosed in the results Meta shared |
| GPT-6.2 Astra | OpenAI | Comparison and review tasks, with overall quality 4% above the November average | 20% below average |
| GPT-6.2 Sol | OpenAI | Above-average quality | 78% below average |
The split is the real story. Meta’s model won on quality, while OpenAI’s two models were pitched on value. For a law firm working through thousands of contracts, a model that is nearly as good at a fraction of the price can win the business even if it does not top the chart.
The results Meta shared did not say what Muse Wonder costs per case. That leaves the most important question for buyers open: how much extra quality is worth paying for.
Also Read: Meta’s Muse Wonder Full Scores 79, Trails Anthropic’s Claude By Two Points
Why Legal Work Is A Hard Test
Legal work punishes small mistakes. A model has to pull the right facts out of long documents, answer questions without inventing anything, draft language that holds up, and edit without changing what a clause means. Muse Wonder led on all four of those task types, which is why Meta is using the result to argue its AI is ready for serious business work.
It also adds a twist to earlier scoring. Fathom previously reported that Muse Wonder trailed Anthropic‘s Claude by two points on an overall score. A win on narrow, high-stakes legal tasks suggests the gap between top models now depends heavily on the job being done.
How Meta Is Pushing Muse Into Work
The legal result caps a busy month for the Muse family. On Nov. 6, Meta brought specialized models trained using Muse Wonder to Muse and Meta AI, saying they give users some of Muse Wonder’s “frontier capabilities” for “complex tasks at work and in everyday life.” Those include Active Memory and Active Focus, which Meta said help the models stay aligned with a user’s goals as tasks unfold.
On Nov. 9, Meta introduced Muse Ember, its smallest and most efficient model, built for “complex reasoning execution and everyday agent work.” It has a 1M-token context window, which means it can keep a very large amount of material in view at once. The same day, Meta brought Muse Ember, Muse Wonder and Muse Spark 1.3 to Google Cloud, so businesses can use them there.
What To Watch Next
Legora publishes its results every month, so December’s numbers will show whether Muse Wonder holds the top spot or OpenAI closes the gap on quality while staying cheaper.
Watch for Meta to share pricing per task. Law firms and corporate legal teams buy on cost as much as accuracy, and a quality lead only turns into contracts if the price makes sense.
The bigger shift is that AI makers now compete on specific jobs, not just general chat. Legal work is one of the first fields where results are tracked task by task, and wins there are likely to shape which models firms sign up for in 2027.
Read Next: Meta Teams Up With Rival Google To Put Its 3 Muse AI Models In The Cloud
