Breaking

Jev Beats Claude Opus 5.5 in 92-Second Tetris Decision Test

September 25, 2026 at 10:17 EDT

AI Agents
Foundation Models

Jev, TypeSafe's decision-only AI model, beat Claude Opus 5.5 in a 92-second real-time Tetris run, clearing 16 lines before topping out against Opus 5.5's 15. The result was narrow on game score but stark on operating metrics: the Jev run was reported as about 9x faster in responses and about 148x lower in estimated API cost.

Sep 25, 2026 · TypeSafe vs Anthropic

Jev beats Opus 5.5 at Tetris

A 92-second run ended with Jev clearing 16 lines, one more than Claude Opus 5.5, while posting a much wider gap on response speed and estimated API cost.

92s

Total Tetris run

Lines cleared by Jev

~9x

Reported faster responses

~148x

Lower estimated API cost

Response speed gap

Shown as a proportional 1x baseline against the reported ~9x Jev response advantage.

Opus 5.5

~9x

Jev

What the win supports

Fast, bounded decision calls can matter more than broad generation when an agent must choose actions repeatedly in real time.

What remains open

Harness design matters: a selector choosing among prepared legal moves is not the same test as a full player reasoning from raw game rules.

The bigger signal

Narrow score, wide operating gap. Jev won by one line, but the reported speed and cost differences point to a broader agent pattern: use decision models for constrained choices and reserve frontier LLMs for open-ended reasoning.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.