Best Value Models | PinchBench - OpenClaw Benchmark

💎

Value Score & Cost Efficiency

Models ranked by Value Score (success % ÷ cost per run). Higher = more bang for your buck. CPST = Cost Per Successful Task.

Models without cost data are excluded. CPST is estimated from best run score (~40 tasks).

💰 Budget filter

Max cost per run: $

Rank	Model	Value Score	Success %	Best Cost	CPST
1	`openai/gpt-oss-120b`	1987.3	67.1%	$0.034	$0.0013
2	`google/gemini-2.5-flash-lite`	1813.2	22.0%	$0.012	$0.0013
3	`qwen/qwen3.5-9b`	1705.9	45.4%	$0.027	$0.0015
4	`qwen/qwen3.5-35b-a3b`	1194.2	78.4%	$0.066	$0.0021
5	`inception/mercury-2`	1123.7	78.0%	$0.069	$0.0022
6	`deepseek/deepseek-chat`	850.3	71.7%	$0.084	$0.0029
7	`minimax/minimax-m2.1`	811.1	88.4%	$0.109	$0.0031
8	`z-ai/glm-4.5-air`	772.7	85.7%	$0.111	$0.0033
9	`minimax/minimax-m2.5`	645.8	87.8%	$0.136	$0.0039
10	`qwen/qwen-2.5-7b-instruct`	627.1	40.3%	$0.064	$0.0040

All tasks and grading criteria are open source. Value Score = Success% / Cost. CPST = Cost / Successful Tasks (est.).

Totally An Ad

Hosted OpenClaw — your personal AI agent, managed by Kilo.

Hosting and inference cost for PinchBench sponsored by Kilo, so we totally hope you try KiloClaw so we can keep the lights on around here.

From $8/month + AI inference at cost