All research

An independent eval of the Paper and Figma MCP servers as design tools for a coding agent, scored on visual quality, consistency, and cost.

I mapped the GPT-5.6 family across task families and difficulty, with an Anthropic comparison, to find the cheapest model that clears your reliability bar.

I ran a controlled eval across six speech-to-text providers, 240 audio clips, and eight content domains to find where voice agents break and which STT model comes out ahead.

Applying the best-performing Parallel configuration from the World Cup eval to inspect the USA vs Belgium matchup as a source-backed research map.

We used Braintrust to evaluate six Parallel web research configurations across 48 World Cup matchups

A Braintrust-native eval comparing GLM-5.2, Opus 4.8, and Sonnet 5 on exact long-context retrieval, cost, and latency.

The full methodology, hypotheses, and findings for Braintrust's GLM-5.2, Opus 4.8, (and Sonnet 5!) long-context retrieval eval.

Call GLM-5.2 directly in playgrounds, prompts, and scorers, with no configuration, through July 31.

We pulled 1,781 real agent traces from Hugging Face into Braintrust, scored every run, and found that the wrapper around your model explains 7× more variation in success than the model itself.

How the cheapest model is not always the cheapest system, and how evals plus control logic reduce cost per resolved request.

Testing whether filesystems and bash provide the optimal abstraction for AI agents through rigorous evaluation.

Learn how aspirational evals can help you figure out when new AI models unlock new product opportunities.

Which one you should ship with, and how to know for sure.

xAI recently announced Grok 4. We put it to the ultimate test.

Faster, more efficient, and highly accurate for real-world applications.