Jev 1 prefill · 6 branches
Ticket (prefill)
Q1
Q2
Q3
Q4
Q5
Q6
→ Intent
→ Order id
→ Amount
→ Damage
→ Refund?
→ Escalate?
The ticket is read once, then each question is answered on its own. No answer can see another.
The realtime decision model
One call.
Whole job.
Rook reads the input and every question together, and returns every decision with its evidence in a single pass. Frontier-level knowledge at realtime speed.
Rook Running…
→ POST /v1/decide
100 tickets · 10 questions each
…
Jev Running…
→ prefill · 100 tickets
…
GPT 6.0 Astra Call 0 / 1,000
#0001 waiting…
Timings are placeholders until measured.
Smarter
[Reasoning benchmark] · accuracy
Rook [X]%
Jev [X]%
Opus 5.5 [X]%
GPT 6.0 Astra [X]%
Faster
[Workload] · decisions per second
One request, 50 decisions
Rook [X]
Jev [X]
Opus 5.5 [X]
GPT 6.0 Astra [X]
Cheaper
[MMLU] · cost per 1k questions · lower is better
Rook $[X]
Jev $[X]
Opus 5.5 $[X]
GPT 6.0 Astra $[X]
Benchmarks, scores and bar lengths are placeholders.
Every answer alone.
Every answer together.
Rook 1 pass
Ticket
Q1–6
Intent
Order id
Amount
Damage
Refund?
Escalate?
The ticket and every question go through together. Each answer is informed by the others.
Our reading of Jev, from public behaviour.
From the blog
All of MMLU in one request.
All 14,042 questions, about 1.5M tokens, sent as one API request. How we ran it, how long it took, and how it compares with other leading models.
Watch it work.
Video 1
Bullet chess vs a leading model
2:40Video 2
Injection capture-the-flag
1:55Video 3
Live fact-check
3:10 Stump
Rook.
Send a job you think it gets wrong. We run it and publish every failure, credited to you.
Every failure we have published →Start building.
Create an account to get an API key and use Rook in the console.
Photo
[Founder name] · @[handle]
[One sentence, in your own words.]