Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

🏠 Back to Blog

Tool School: Benchmarking 101 (How To Read AI Model Report Cards)

Authors: Hannah Stulberg and Akshat Khandelwal | Published: 2026-03-31

Learn to read the scorecard behind every AI model launch. Core thesis: “The best model is the one that does the work you need, at a price you can justify, on benchmarks you understand.”

Benchmarks 101

Analogy: Benchmark scores are like SAT scores – a high SAT score doesn’t make someone a great colleague. Each measures one narrow skill.

Three Scoring Systems

SystemHow it worksNotes
Pass@1One attempt per questionClosest to actual user experience. Most commonly reported
Pass@kMultiple attempts (8 or 16); any correct = passAlways higher than pass@1. Some companies report this without clear labeling
Elo ratingHead-to-head ranking (like chess)Chatbot Arena uses this. Measures preference, not correctness

Four Benchmark Categories

  1. Knowledge & Reasoning – academic/professional knowledge (MMLU, GPQA Diamond, HLE, GDPval)
  2. Coding – writing working code (SWE-bench Verified, SWE-bench Pro, SWE-Lancer)
  3. Reasoning – novel problem-solving (ARC-AGI-2, ARC-AGI-3)
  4. Agentic & Computer Use – sustained multi-step professional work (OSWorld, tau2-bench, BigLaw Bench, Terminal-Bench 2.0)

Key Benchmarks

BenchmarkWhat it testsCurrent status
MMLU16K questions, 57 subjectsSaturated (88-93%)
GPQA Diamond448 PhD-level science questionsApproaching saturation (91-94%)
HLE2,500 expert questions, 100+ subjects (published in Nature)Still differentiating (40-45%)
SWE-bench Verified500 real GitHub issuesContaminated – every frontier model trained on solutions
SWE-bench Pro1,900 tasks with private test casesCurrent best coding benchmark
SWE-Lancer1,400+ real Upwork tasks ($50 to $32K)80% SWE-bench = only 26% SWE-Lancer
ARC-AGI-3Interactive game environments, no instructionsWidest human-AI gap (humans 100%, best AI 0.37%)
ARC-AGI-2Visual pattern puzzlesScores still spread (53-77%). Measures cost alongside accuracy
OSWorldReal computer tasks in OS environmentsGPT-5.4 surpassed human performance (75% vs 72.4%)

Why Benchmarks Expire

  1. Saturation – scores converge, spread disappears (MMLU lasted ~3 years, GPQA ~2)
  2. Contamination – test answers leak into training data (SWE-bench Verified confirmed contaminated)
  3. Goodhart’s Law – when a measure becomes a target, it ceases to be a good measure

Three Questions for Any Benchmark

  1. How long has the test been public?
  2. Who created it?
  3. Are the scores still spread out?

Trust Tiers

  • High relevance: ARC-AGI-3, ARC-AGI-2, HLE, SWE-bench Pro, OSWorld, Terminal-Bench 2.0
  • Moderate: GPQA Diamond, tau2-bench, BigLaw Bench, GDPval, MMLU Pro
  • Low: MMLU, SWE-bench Verified, HumanEval

Formula: independent creation + wide score spread + contamination resistance = high relevance

Head-to-Head (as of article date)

StrengthLeader
Academic reasoningGemini 3.1 Pro
Professional tasksGPT-5.4
Enterprise agentic workOpus 4.6

Top 5 models separated by just 19 Elo points in Chatbot Arena.

Cost

  • API pricing: input tokens vs output tokens, per million
  • Same benchmark scores, wildly different prices: DeepSeek R1 at $2.19/M output vs $15 (GPT-5.4) vs $25 (Opus 4.6)
  • Sonnet 4.6 delivers ~95% of Opus 4.6 performance at 60% of the cost
  • For consumer subscriptions ($20/month), cost shows up as rate limits, not per-token

Four Things to Watch in Any Launch

  1. What benchmarks are missing from the scorecard? (selective disclosure)
  2. What baseline is the company comparing against? (may be outdated)
  3. How does Chatbot Arena compare to the benchmark story?
  4. How transparent is the methodology?

Independent Sources for Checking Scores

  • Chatbot Arena, Artificial Analysis, ARC Prize Leaderboard, Scale AI SWE-bench Pro Leaderboard, Stanford CRFM Foundation Model Transparency Index