Terminal-Bench — cost vs performance

Best current row per canonical model in the group “accuracy · variant=hard · evaluator=Artificial Analysis” (317 models) against context window (tokens). 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 (16 models)

Frontier models 16

Pareto frontier
Model	accuracy on Terminal-Bench	Rank	context window (tokens)	Provider	Trust
gpt-5.6-solOpenAI · Closed	65.9%	1	1.05M	—	Independent
Claude Fable 5Anthropic · Closed	62.9%	2	1M	—	Independent
gpt-5.3-codexOpenAI · Closed	53.0%	9	400K	—	Independent
KAT-Coder-Pro V2Kwaipilot · Closed	49.2%	14	262.1K	—	Independent
Claude Opus 4.5Anthropic · Closed	47.0%	16	200K	—	Independent
DeepSeek V3DeepSeek · Open weights	35.6%	54	163.8K	—	Independent
DeepSeek V3.2DeepSeek · Open weights	35.6%	54	163.8K	—	Independent
deepseek-v3-2-specialeDeepSeek · Open weights	34.9%	59	128K	—	Independent
magistral-mediumMistral AI · Closed	9.09%	172	40K	—	Independent
qwen3-30b-a3b-instructAlibaba Group · Open weights	6.82%	186	32.8K	—	Independent
Phi 4Microsoft · Open weights	3.79%	222	16.4K	—	Independent
R1 Distill Llama 70BDeepSeek · Open weights	1.52%	256	8.19K	—	Independent
molmo-7b-dAllen Institute for AI · Open weights	0%	279	4.1K	—	Independent
olmo-2-32bAllen Institute for AI · Open weights	0%	279	4.1K	—	Independent
olmo-2-7bAllen Institute for AI · Open weights	0%	279	4.1K	—	Independent
phi-3-miniMicrosoft · Open weights	0%	279	4.1K	—	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.