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.

DeepSeek V4 FlashDeepSeek V4.1 FlashDeepSeek V4 Pro Add a modelChat with these 3

AttributeHide identical rows	DeepSeek V4 Flashdeepseek/deepseek-v4-flash	DeepSeek V4.1 Flashdeepseek/deepseek-v4.1-flash	DeepSeek V4 Prodeepseek/deepseek-v4-pro
Specs
Context window	1M	1M	1M
Max output	384K	384K	384K
Input modalities	Text	TextImage	Text
Output modalities	Text	Text	Text
Capabilities
Streaming	Supported	Supported	Supported
Tool calling	Supported	Supported	Supported
Structured output (JSON)	Supported	Supported	Supported
Image input	Not supported	Supported	Not supported
Thinking mode	Supported	Supported	Supported
Price / 1M tokens
Input	$ 0.30	$ 0.30	$ 1.35
Output	$ 1.20	$ 1.20	$ 4.05
Cache read	$ 0.0060	$ 0.0060	$ 0.045
Cache write	$ 0.30	$ 0.30	$ 1.35
What it is good at
Best for	Cost-sensitive coding agents High-volume tool workflows	Terminal and coding agents Agents that read screenshots and charts Hard knowledge and research High-throughput production traffic	Complex long-horizon agents Research and hard reasoning
Published benchmarks
Terminal-Bench 2.1pass_ratesetups differ by: reasoning	—	90.6%(setup not fully disclosed)	78.7%effort max(setup not fully disclosed)
Terminal-Bench 4.0pass_rate	—	31.2%(setup not fully disclosed)	—
Agents' Last Examscore	—	31.8%(setup not fully disclosed)	—
AutomationBenchscore	—	54.8%(setup not fully disclosed)	—
DeepSWE 1.1resolved_rate	—	74.2%(setup not fully disclosed)	—
NL2Reposcore	—	65.4%(setup not fully disclosed)	—
SciCodescore	—	—	49.2%effort max(setup not fully disclosed)
CyberGymscore	—	88.1%(setup not fully disclosed)	—
GDPval-AA v2elo	—	—	1590.3effort max(setup not fully disclosed)
AA-LCRscore	—	—	75.3%effort max(setup not fully disclosed)
MathArena Apexscore	—	65.6%(setup not fully disclosed)	—
Chartography (with tools)score	—	78.9%(setup not fully disclosed)	—
AA Intelligence Index v4.1.1score	—	—	53effort max(setup not fully disclosed)
Humanity's Last Examaccuracysetups differ by: reasoning	—	36.8%(setup not fully disclosed)	39.3%effort max(setup not fully disclosed)
Humanity's Last Exam (with tools)accuracy	—	63.9%(setup not fully disclosed)	—
CritPtscore	—	—	18%effort max(setup not fully disclosed)
GPQA Diamondaccuracysetups differ by: reasoning	—	90.9%(setup not fully disclosed)	92.8%effort max(setup not fully disclosed)
tau3-Bankingscore	—	—	39.6%effort max(setup not fully disclosed)