Pass · production benchmark

400 real messages, four people, one compiler — with memory.

Four user profiles ran against production Pass — every request compiled by the small model: route, goal, language, what to pull. New since last time: a standing user profile, write-time-reconciled memory, hybrid retrieval with a cross-encoder judge, and a fourth character built to test all of it. All 400 messages also went to Anthropic's Opus 5, and an independent judge scored both.

2026-08-26400/400 delivered · 0 errors10 languagesmemory · knowledge base · web
100%
of decisions made by the model — route, goal, language and every pull, one call per request
86.7
avg quality (0–100 rubric) across Alice, Bob and Claude vs 66.9 for Opus 5 — 130% of the frontier
2.1×
faster — 7.4 s median answer across the four vs 15.7 s for Opus
~80×
cheaper inference — $0.00033/msg average vs $0.0237 for Opus

Four user profiles

Personal — Alice
Everyday life: trips, recipes, family messages, explanations for kids, quick translations.

Speed — median 9.4 s vs Opus 15.7 s · 1.7× faster
Cost — $0.000434/msg inference vs Opus $0.0196 · 45× cheaper
Quality — 90.7 vs Opus 68.6 · 132%
Business — Bob
Work: client emails, quarterly agendas, risk analysis, hiring trade-offs, status reports.

Speed — median 8.7 s vs Opus 20.5 s · 2.4× faster
Cost — $0.000376/msg inference vs Opus $0.0281 · 75× cheaper
Quality — 87.4 vs Opus 65.3 · 134%
Coder — Claude
Engineering: algorithms, SQL with window functions, API design reviews, decorators and complexity.

Speed — median 6.2 s vs Opus 15.7 s · 2.5× faster
Cost — $0.000403/msg inference vs Opus $0.0302 · 75× cheaper
Quality — 82.0 vs Opus 66.7 · 123%
Memory — Dobby
A 78-year-old who forgets. He asks short and vague — "what's up with the pressure?" — and the assistant must know from his life whether that means his blood pressure, his boiler, or the weather. His documents sit in a knowledge base; his facts live in Pass's memory; nothing is in the message.

Speed — median 6.3 s vs Opus 11.3 s · 1.8× faster
Cost — $0.000097/msg inference vs Opus $0.0169 · 174× cheaper
Quality — 61.2 vs Opus 23.5, informed judge · 260%

Judging Dobby honestly

For Alice, Bob and Claude the judge (gemini-3.6-flash) scored blind — each answer alone against a 0–100 rubric: correctness, then completeness, then clarity. For Dobby a blind judge is structurally wrong: it cannot know that "the pressure" is a Vaillant boiler reading 1.2–1.5 bar, so it rates a polite "I can't access that, try an app" at 95. The Dobby judge was therefore given the ground truth of his life — city, medication, boiler, prices, appointments — and asked whether the answer serves this user. With facts in hand: Pass 61.2, Opus 23.5. A model without memory cannot answer "what's the weather?" with the weather in Faro.

What the compiler decides

Every request produces a Job on the receipt: the route, the goal in the compiler's own words, the language of the answer, and the pulls — knowledge base, memory, standing profile, web, rerank — in order. Reads the compiler orders are judged by a cross-encoder before anything enters the prompt; a read that finds nothing is said out loud, never papered over; an emptied read may escalate to the web once, and the receipt names it. The assistant's own answers are never written to memory — only the user's words are.

Method

400 unique prompts — 100 per profile, 10 per language — against production, cache off. All 400 went to Opus 5 with identical prompts (max_tokens 1200) over OpenRouter. Costs are inference only, per message: Pass from token usage at registry prices (its serving models are small — the compiler routes most traffic to them), Opus from its own usage; Pass's $0.01 service fee is separate from these numbers. Answers truncated to 1400 characters for the judge. Previous benchmarks: 2026-08-25 · 2026-08-19.