Terminal-Bench — cost vs performance

Best current row per canonical model in the group “accuracy · variant=hard · evaluator=Artificial Analysis” (317 models) against ESTIMATED memory at 4-bit, 8K context (GB). The dashed line is the Pareto frontier: no model is both better and cheaper than a point on it.

← Leaderboard

XOutput price Input price Parameters Context Memory (est.)Scalelog linearBubblecontext params none

open / restricted weights closed Pareto frontier (10 models)bubble = context windowEstimated

Frontier models 10

Pareto frontier
Model	accuracy on Terminal-Bench	Rank	ESTIMATED memory at 4-bit, 8K context (GB)	Provider	Trust
Z.ai GLM 5.2Z.ai (Zhipu AI) · Open weights	50.8%	12	433.7 GBest.	—	Independent
MiniMax-M3MiniMax · Open weights	42.4%	30	246.1 GBest.	—	Independent
MiniMax M2.7MiniMax · Open weights	39.4%	39	132 GBest.	—	Independent
Gemma 4 31BGoogle · Open weights	36.4%	51	18.5 GBest.	—	Independent
Qwen3.6 27BQwen · Open weights	34.9%	59	16.5 GBest.	—	Independent
Gemma 4 26B A4BGoogle · Open weights	25%	100	15.3 GBest.	—	Independent
Qwen3.5-9BQwen · Open weights	24.2%	105	6.1 GBest.	—	Independent
Qwen3.5-4BQwen · Open weights	18.2%	127	3.2 GBest.	—	Independent
granite-4.1-3bIBM · Open weights	2.27%	240	2.5 GBest.	—	Independent
Qwen3-0.6BQwen · Open weights	0%	279	0.9 GBest.	—	Independent

Methodology. Points are the best current row per canonical model in comparability group 'accuracy · variant=hard · evaluator=Artificial Analysis'. Price = cheapest current offer across providers (the provider shown). Pareto frontier maximises the score and minimises x; exact ties are all kept. memory_estimate is an estimate (see /methodology). Only the selected comparability group is plotted; points under other configurations are not mixed in. Price = cheapest current offer across providers at the time of the last crawl. Nothing is estimated except where marked.