Compare models

Context, limits, capabilities, price and use cases side by side — up to 4 models at a time. The URL holds your selection, so you can share it.

GPT-5.6 LunaGPT-5.6 TerraGPT-5.4 Nano Add a modelChat with these 3

AttributeHide identical rows	GPT-5.6 Lunaopenai/gpt-5.6-luna	GPT-5.6 Terraopenai/gpt-5.6-terra	GPT-5.4 Nanoopenai/gpt-5.4-nano
Specs
Context window	1M	1M	400K
Max output	128K	128K	128K
Input modalities	TextImage	TextImage	TextImage
Output modalities	Text	Text	Text
Capabilities
Streaming	Supported	Supported	Supported
Tool calling	Supported	Supported	Supported
Structured output (JSON)	Supported	Supported	Supported
Image input	Supported	Supported	Supported
Thinking mode	Supported	Supported	Supported
Price / 1M tokens
Input	$ 0.20>= 272K$ 0.40	$ 2.00>= 272K$ 4.00	$ 0.20
Output	$ 1.20>= 272K$ 1.80	$ 12.00>= 272K$ 18.00	$ 1.25
Cache read	$ 0.020>= 272K$ 0.040	$ 0.20>= 272K$ 0.40	$ 0.020
Cache write	$ 0.25>= 272K$ 0.50	$ 2.50>= 272K$ 5.00	$ 0.20
What it is good at
Best for	High-volume agent execution Routine implementation and subagents	Everyday knowledge work Cost-aware agentic coding	Classification and extraction Ranking and simple subagents
Published benchmarks
Terminal-Bench 2.0pass_rate	—	—	46.3%effort xhigh(setup not fully disclosed)
Terminal-Bench 2.1pass_rate	84.7%effort max(setup not fully disclosed)	87.4%effort max(setup not fully disclosed)	—
BrowseComp (with tools)accuracy	83.3%effort max(setup not fully disclosed)	87.5%effort max(setup not fully disclosed)	—
Agents' Last Examscore	50.3%effort max(setup not fully disclosed)	50.4%effort max(setup not fully disclosed)	—
SWE-Bench Pro (Public)pass_ratesetups differ by: reasoning	62.7%effort max(setup not fully disclosed)	63.4%effort max(setup not fully disclosed)	52.4%effort xhigh(setup not fully disclosed)
OSWorld 2.0success_rate	45.6%effort max(setup not fully disclosed)	50.2%effort max(setup not fully disclosed)	—
OSWorld-Verifiedsuccess_rate	—	—	39%effort xhigh(setup not fully disclosed)
OpenAI MRCR v2 8-needle 128K–256Kaccuracy	—	—	33.1%effort xhigh(setup not fully disclosed)
OpenAI MRCR v2 8-needle 512K–1Maccuracy	41.3%effort max(setup not fully disclosed)	72.5%effort max(setup not fully disclosed)	—
MMMU Pro (no tools)accuracy	—	—	66.1%effort xhigh(setup not fully disclosed)
MMMU Pro (with tools)accuracy	79.5%effort max(setup not fully disclosed)	82%effort max(setup not fully disclosed)	—
Humanity's Last Exam Full (with tools)accuracy	—	—	37.7%effort xhigh(setup not fully disclosed)
GPQA Diamondaccuracysetups differ by: reasoning	92.3%effort max(setup not fully disclosed)	92.9%effort max(setup not fully disclosed)	82.8%effort xhigh(setup not fully disclosed)
Toolathlonscoresetups differ by: reasoning	53.4%effort max(setup not fully disclosed)	53.1%effort max(setup not fully disclosed)	35.5%effort xhigh(setup not fully disclosed)