Harnesses

A harness is the program that drives a model through a coding task: Claude Code, Codex. Each is ranked by its best run on one source, with the model behind it named, because part of a harness’s score is earned by the model it drives.

Harness Best pairing AA Cost per task

1 Claude Code Claude Sonnet 5.5 · max effort 0.684 $14.19

All 9 runs
- Claude Sonnet 5.5 · max effort SOTA 0.684 $14.19
- Claude Opus 5.5 · max effort SOTA 0.660 $13.04
- Claude Sonnet 5.5 · xhigh effort KILLED 0.629 $3.33
- Claude Fable 5.1 · max effort · unknown fallback KILLED 0.622 $12.39
- Claude Opus 5 · max effort KILLED 0.597 $10.79
- Claude Sonnet 5.5 · high effort KILLED 0.550 $1.24
- Claude Sonnet 5.5 · medium effort LOW-COST 0.459 $0.62
- Qwen 3.8 Max KILLED 0.433 $3.48
- Claude Sonnet 5.5 · low effort LOW-COST 0.421 $0.48
2 Antigravity CLI Gemini 4 Argon 0.638 $5.84

All 1 runs
- Gemini 4 Argon SOTA 0.638 $5.84
3 Codex GPT 6.1 Sol · xhigh effort 0.629 $1.04

All 12 runs
- GPT 6.1 Sol · xhigh effort Executioner 0.629 $1.04
- GPT 6 Astra · max effort KILLED 0.616 $7.47
- GPT 6.1 Sol · medium effort LOW-COST 0.614 $0.70
- GPT 6.1 Sol · high effort LOW-COST 0.601 $0.89
- GPT 6.1 Sol · max effort KILLED 0.601 $1.55
- GPT 6.1 Sol · low effort LOW-COST 0.572 $0.50
- GPT 6 Sol · max effort KILLED 0.567 $2.99
- GPT 5.6 Sol · max effort KILLED 0.546 $6.35
- GPT 5.6 Luna · max effort LOW-COST 0.432 $0.44
- DeepSeek V4 Pro · max effort · 0813 LOW-COST 0.431 $0.24
- GPT 6 Luna · max effort LOW-COST 0.411 $0.18
- DeepSeek V4 Flash · max effort · 0731 LOW-COST 0.387 $0.09
4 Devin Fusion CLI Claude Fable 5.1 XHigh + SWE-2 Medium 0.617 $7.90

All 2 runs
- Claude Fable 5.1 XHigh + SWE-2 Medium KILLED 0.617 $7.90
- GPT-6 Astra XHigh + SWE-2 Medium KILLED 0.589 $4.54
5 Grok Build Grok 4.7 · xhigh effort 0.563 $8.82

All 2 runs
- Grok 4.7 · xhigh effort KILLED 0.563 $8.82
- Grok 4.6 · xhigh effort KILLED 0.470 $3.57
6 Muse Code Muse Spark 1.3 · max effort 0.543 $3.98

All 1 runs
- Muse Spark 1.3 · max effort KILLED 0.543 $3.98
7 Opencode GLM 5.3 0.536 $4.24

All 1 runs
- GLM 5.3 KILLED 0.536 $4.24
8 Kimi Code CLI Kimi K3 0.519 $5.05

All 1 runs
- Kimi K3 KILLED 0.519 $5.05
9 Muse Code 1.0.2 RC Muse Spark 1.3 · xhigh effort 0.483 $3.47

All 1 runs
- Muse Spark 1.3 · xhigh effort KILLED 0.483 $3.47
10 Antigravity SDK v0.1.12 Gemini 3.8 Flash · high effort 0.419 $2.47

All 1 runs
- Gemini 3.8 Flash · high effort KILLED 0.419 $2.47

Published by: Artificial Analysis
Licence: not stated
Updates: continuous
Fetched: 2026-10-04
Entries: 31 · 2 unmatched

The kill line, over pairings

Every harness-and-model pair the source priced, placed by cost per task and agent index. A harness charges nothing itself; the model it drives runs up most of the bill.