← All models
Opus 5
claude-opus-5
Head to head
Measured
Run locally in Bench Arena. Tokens reconciled on the harness log, cost recomputed from them.
| Prompt | Status | Cost | Duration | Output |
|---|---|---|---|---|
| Jelly Jungle — 3D jungle run | Timed out | $20.9 | 2h 00m | 218k |
| Jelly Jungle — 3D jungle run | Failed | $21.3 | 1h 09m | 260k |
| Call of Duty in the browser — Claude Code vs Codex CLI on one prompt | Interrupted | $18.6 | 1h 29m | 159k |
| Wobble Rush — 3D obstacle course | Interrupted | $9.39 | 23m 15s | 103k |
| Wobble Rush — 3D obstacle course | Completed | $10.5 | 34m 41s | 124k |
Reported
Announced by the author of the source. We did not run it and cannot verify these numbers.
| Prompt | Status | Cost | Duration | Output |
|---|---|---|---|---|
| Kage — a Three.js night walk through a Kyoto mountain temple | Completed | not measured | not measured | not measured |
| Descent — Opus 5 shipped a cel-shaded downhill BMX racer from one prompt | Completed | $863.4 | 23h 28m | 5400k |
| Counter-Strike in the browser — Kimi K3, Fable 5, Opus 5 | Completed | not measured | not measured | not measured |
| Ink Tide — Opus 5 shipped a cel-shaded boat racer from one prompt | Completed | $423 | not measured | not measured |
| Glass aquarium burst — cracked panel, hydrostatic jet | Completed | not measured | not measured | not measured |
| Halo — Blood Gulch 5v5 in the browser | Completed | not measured | 12h 00m | not measured |
| The Long Silence — Opus 5 built a space sim in the browser over 24 hours | Interrupted | not measured | 24h 00m | not measured |
| 3D destruction physics — tornado, wrecking ball, truss bridge | Completed | $1.40 | not measured | 56k |
| Counter-Strike 2 — clone it in the browser | Completed | not measured | 37h 00m | not measured |