· via dev.to (home feed)
Meta's Muse Wonder tops Legora legal AI benchmark, beating OpenAI on quality
Meta's Muse Wonder scored highest on overall quality in Legora's November legal benchmark, leading four task types, while OpenAI's GPT-6.2 models won on cost per case.

What happened
Meta's Muse Wonder model has taken the top overall quality score in a monthly evaluation of how AI handles real legal work, finishing ahead of rival models from OpenAI. The ranking comes from Legora, a legal AI platform used by law firms, which publishes monthly results from its BAR benchmark measuring how models perform on practical legal tasks inside its product, according to Fathom News.
Muse Wonder led four of the task categories tracked by the benchmark: analysis and answering, extraction, drafting, and editing. In a post on X, Meta described legal work as "a tough test for AI" and called the outcome "a strong showing on the kinds of complex tasks businesses need help with."
OpenAI counters on cost
OpenAI's models did not leave empty-handed. GPT-6.2 Astra led the comparison and review category and posted overall quality 4% above the November average, at a cost per case 20% below average. Its sibling, GPT-6.2 Sol, delivered above-average quality at 78% below the average cost per case.
That split is the most consequential part of the result. Meta won on quality; OpenAI won on value. A law firm moving through thousands of contracts may reasonably pick a model that is nearly as accurate at a small fraction of the price, even if it does not sit at the top of the chart.
The catch is that the results Meta shared did not disclose what Muse Wonder costs per case, according to Fathom News. That leaves the central question for buyers unanswered: how much premium the quality lead is worth.
Rankings shift with the job
The result also complicates earlier scoring. Fathom News previously reported that Muse Wonder scored 79 on an overall measure, two points behind Anthropic's Claude. Winning on narrow, high-stakes legal tasks suggests the ordering of top models now depends heavily on what the model is actually being asked to do, rather than on any single general-purpose score.
Legal work is a demanding test because errors carry real consequences. A model has to pull the correct facts out of long documents, answer questions without inventing anything, draft language that holds up, and edit text without altering what a clause means. Muse Wonder led all four of those task types in the November results.
Meta's wider Muse rollout
The legal result caps an active stretch for Meta's model family. On November 6, the company brought specialized models trained using Muse Wonder to Muse and Meta AI, including features called Active Memory and Active Focus, which Meta says help models stay aligned with a user's goals as a task unfolds.
On November 9, Meta introduced Muse Ember, its smallest and most efficient model, built for complex reasoning execution and everyday agent work, with a 1M-token context window that keeps very large amounts of material in view at once. The same day, Meta made Muse Ember, Muse Wonder, and Muse Spark 1.3 available on Google Cloud for business customers.
What to watch
Legora publishes its benchmark monthly, so December's figures will show whether Muse Wonder keeps the top quality spot or OpenAI closes the gap while staying cheaper. Pricing disclosure is the other open item: Fathom News notes that legal buyers weigh cost as much as accuracy, and a quality lead only converts into contracts if the price makes sense.
Why it matters
Professional services are becoming the proving ground for AI adoption, and legal is one of the first fields where model performance is tracked task by task on real work rather than on abstract tests. This benchmark shows that the competition has already fragmented: one vendor leads on quality, another on unit economics, and buyers will make procurement decisions on that trade-off. As AI makers increasingly compete on specific jobs rather than general chat, results like Legora's are likely to shape which models law firms and corporate legal teams sign up for in 2027.
- #meta
- #openai
- #legal-ai
- #benchmarks
- #llm