← All writing

Tag

#benchmarks

  1. Agent Coordination: Why AI Agent Teams Talk Too Much

    Agent coordination tested: Bespoke Labs' MAC Bench found eight AI agents beat one, but eight agents that never talked did just as well. Here is why.

  2. Strict Pass Rate: Stripe Evals Show the Last-Mile Gap

    Stripe evals found coding agents meet 78.1% of payment criteria on average, yet the strict pass rate is just 29.4%. Here is what that gap means for you.

  3. Gemini 4 Neon vs GPT-6.2 and Claude: Price and Speed

    Gemini 4 Neon costs $3/$15 per million tokens. How it compares with GPT-6.2 Astra, Sol, Claude Fable 5.2 and Opus 5.5 on price, scores and cost per task.