HarnessTax — SWE-bench Lite

Shown as HarnessTax (Arena) publishes it — their metrics, their models. A model name opens its page here; the arrow opens the original board.

Published by

HarnessTax (Arena)

Licence

not stated

Updates

frozen

Fetched

2026-10-05

Entries

21

Filter modelsOrder bySWE-bench LiteCost per rolloutMatched models only

21 scored

#	Model	SWE-bench Lite ↑	Cost per rollout ↓
1	Claude Fable 5Claude CodeAnthropic	97.8%93.3–100.0	$1.33
2	Claude Fable 5CodexAnthropic	96.7%91.1–100.0	$0.89
3	Claude Fable 5PiAnthropic	96.7%91.1–100.0	$0.67
4	Claude Opus 4.8CodexAnthropic	88.9%77.8–97.8	$0.69
5	Claude Opus 4.8Claude CodeAnthropic	86.7%75.6–95.6	$0.98
6	Claude Opus 4.8PiAnthropic	82.2%71.1–92.2	$0.47
7	GPT 5.6 SolClaude CodeOpenAI	77.8%63.3–90.0	$1.54
8	Kimi K3Claude CodeKimi	76.7%62.2–88.9	$0.78
9	GPT 5.6 SolPiOpenAI	74.4%58.9–88.9	$0.44
10	Kimi K3CodexKimi	74.4%60.0–87.8	$0.85
11	GPT 5.6 SolCodexOpenAI	73.3%57.8–87.8	$0.56
12	Kimi K3PiKimi	72.2%57.8–85.6	$0.46
13	Claude Sonnet 4.6CodexAnthropic	68.9%53.3–83.3	$0.74
14	Claude Sonnet 4.6Claude CodeAnthropic	66.7%51.1–81.1	$0.67
15	Claude Sonnet 4.6PiAnthropic	64.4%47.8–80.0	$0.68
16	Claude Haiku 4.5PiAnthropic	60.0%43.3–75.6	$0.37
17	Claude Haiku 4.5CodexAnthropic	57.8%41.1–74.4	$0.39
18	GPT 5.6 LunaClaude CodeOpenAI	55.6%38.9–72.2	$0.15
19	GPT 5.6 LunaCodexOpenAI	55.6%37.8–72.2	$0.04
20	GPT 5.6 LunaPiOpenAI	53.3%35.6–70.0	$0.03
21	Claude Haiku 4.5Claude CodeAnthropic	52.2%36.7–67.8	$0.43

What each column measures

SWE-bench Lite

Higher is better · Agentic coding

Cost per rollout

Lower is better