Tessl: Jev is 13.6x faster and 2.7x cheaper than GPT Luna 6 as an LLM code verifier
aheineike · x · 2026-09-24
Tessl benchmarked TypeSafe's new decision model Jev as a drop-in replacement for the LLM "verifiers" (linting with reasoning) it runs on production code, comparing against its current judge GPT Luna 6:
- Scope: 6 projects, 2,725 verifier-file pairs, gateway cache bypassed so every call was a fresh generation
- Jev: $0.65 / 32s vs GPT Luna 6's $1.74 / 436.5s — 13.6x faster, 2.7x cheaper ($0.24 vs $0.64 per 1,000 targets)
- Against the previous Luna 5.6, Jev was 6.6x cheaper and 10x faster; Luna 6 got cheaper but slower, narrowing Jev's cost edge
- Methodology, model details, and commands are published so you can rerun it on your own repo
A practical pattern worth stealing: use a small decision model for large-scale code review instead of a flagship LLM.
Related event: Jev Beats GPT Luna as Code Verifier, 13.6x Faster(2 posts)→
More from coding & agent
- Anthropic made claude.ai 3x faster in two weeks — and shares how, prompts included — bcherny · 2026-09-24
- Walkthrough: Blender MCP plus FLORA for precise product turntable animations — round · 2026-09-24
- New video series on agent evals kicks off with episode 1: how to read traces — doesdatmaksense · 2026-09-24
- Open-source agent lead-enrichment stack Jev+Treg undercuts Clay at $0.029 per verified lead — cneuralnetwork · 2026-09-24
- uv author: metadata-free lockfiles still detect stale resolutions — charliermarsh · 2026-09-24
- Running Claude vs Codex debates in a shared doc works, but it's slow and expensive — kshitizsriv · 2026-09-24