NewsNew models, Next Concept Models and AI agents · coming soonThe VerdictHead-to-head tests you can rerun on LMArena · coming soonFuture with AIEssays on where AI is going · coming soonEverything in Read →
Coming soonThe first News, Verdicts and essays are on the way.Get them first in AI into Production, our weekly newsletter.
Learn
Practical guides to using AI models well
GuidesHow-to pieces on cost, routing and usage · coming soonLLM glossaryPlain definitions of the terms we use · coming soonBYOK 101Using your own API keys, safely · coming soonProvider and lab reviewsWhere to run each model, and what it costs · coming soon
LanguageTranslate this page
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
NewsNew models, Next Concept Models and AI agents · coming soonThe VerdictHead-to-head tests you can rerun on LMArena · coming soonFuture with AIEssays on where AI is going · coming soonEverything in Read →
Coming soonThe first News, Verdicts and essays are on the way.Get them first in AI into Production, our weekly newsletter.
Learn
Practical guides to using AI models well
Guides7How-to pieces on cost, routing and usageLLM glossaryPlain definitions of the terms we use · coming soonBYOK 101Using your own API keys, safely · coming soonProvider and lab reviewsWhere to run each model, and what it costs · coming soon
Legal Agent Benchmark mean criterion-pass rate — Anthropic harness. Every model the catalog carries a published LAB criterion-pass (Anthropic harness) value for, ranked by that value.
Legal Agent Benchmark mean criterion-pass rate — Anthropic harness. Every model the catalog carries a published LAB criterion-pass (Anthropic harness) value for, ranked by that value.
About this benchmark
Mean fraction of expert-written rubric criteria passed across legal-agent tasks. One model in the ModelChoice catalog has a published value.
When it's a good referenceUse it when choosing a model for multi-step work with tools, where the model acts rather than only answers. Check how the agent was set up before comparing two rows.
FrontierThe highest score on this leaderboard.Claude Opus 593.74 on LAB criterion-pass (Anthropic harness).
Daily driverThe cheapest best-value model scoring at least 95% of the frontier pick's score, or, if none does, the highest-scoring best-value model.No model qualifies yet.
EfficientThe best-value model with the most score per dollar among those scoring at least 80% of the frontier pick's score, or among all best-value models if none does.No model qualifies yet.
9095
Quality versus cost
Each point is a model with a published LAB criterion-pass (Anthropic harness) value and complete token prices, on the 75% input / 25% output blend. Frontier models are spaced evenly along the cost axis.
LAB criterion-pass (Anthropic harness) ranking
1 model with a published LAB criterion-pass (Anthropic harness) value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.
Models ranked by their published Legal Agent Benchmark mean criterion-pass rate — Anthropic harness value
Rows are ordered by the value Claude Opus 5 System Card published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.
About LAB criterion-pass (Anthropic harness)
What the benchmark measures and how its value is published, from the source’s own record.
What is LAB criterion-pass (Anthropic harness)?
LAB criterion-pass (Anthropic harness) (Legal Agent Benchmark mean criterion-pass rate — Anthropic harness) is an agentic benchmark published by Claude Opus 5 System Card. Mean fraction of expert-written rubric criteria passed across legal-agent tasks. Its evaluation set is described by the source as 1,235 legal-agent tasks. The source rates its difficulty as professional legal work.
How is LAB criterion-pass (Anthropic harness) scored?
Each model's result is published by Claude Opus 5 System Card as mean criterion-pass rate. ModelChoice prints that value to 2 decimal places, exactly as published, and never converts or rescales it. This page ranks the 1 model that carry a published value, highest value first; a model the source never scored is left out rather than ranked at zero. The source does not state whether a higher value is the better result, so neither does this page.
MonoMind is an AI transformation partner for businesses.
We are a digital agency and AI consulting studio. We help companies move from “we should be using AI” to AI that works inside their business and produces results.
What we do
AI Strategy
We start with your business, not the technology. We find where AI can reduce cost, save effort, or create new revenue, then build a clear roadmap you can act on.
Build & Deploy
We design, build, and launch the websites, platforms, and AI-powered tools in that roadmap. Everything we build is meant to go live and be used.
Marketing & Growth
A good product still needs to reach people. We turn your digital presence into a growth channel through content, search, and campaigns built for your audience.
How we work
One team, start to finish. The people who shape your strategy also deliver it. Nothing gets lost between planning and launch.
Priced by value. We charge based on the outcome we deliver, not the effort it takes. Our incentive is your result.
Local understanding. We know the languages, platforms, and business culture of the markets we serve, so our work fits how your customers actually behave.
Our belief
AI should make a business simpler and stronger, not more complicated. Our name reflects that: Mono for focus, Mind for intelligence. We focus on the few things that matter most, and we use AI to get them done well.
Let’s talk about what AI can do for your business. daren@monomind.one · https://monomind.one
ModelChoiceby MonoMind AI Lab
Evidence for practical AI runtime and cost decisions.
Every figure in this document comes from a cited upstream source and carries the date it was read. A value no source published is shown as unavailable, never as zero. The live page is updated as the sources change.