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.

Kimi K2.5Kimi K2.6 Add a modelChat with these 2

AttributeHide identical rows	Kimi K2.5moonshot/kimi-k2.5	Kimi K2.6moonshot/kimi-k2.6
Specs
Context window	262K	262K
Max output	262K	262K
Input modalities	TextImageVideo	TextImageVideo
Output modalities	Text	Text
Capabilities
Streaming	Supported	Supported
Tool calling	Supported	Supported
Structured output (JSON)	Supported	Supported
Image input	Supported	Supported
Thinking mode	Supported	Supported
Price / 1M tokens
Input	$ 0.62	$ 1.00
Output	$ 3.30	$ 4.16
Cache read	$ 0.11	$ 0.18
Cache write	$ 0.62	$ 1.00
What it is good at
Best for	Visual-to-code Agentic search	Agentic coding Tool-assisted research Multimodal workflows
Published benchmarks
Terminal-Bench 2.0pass_rate	—	66.7%Terminus 2(setup not fully disclosed)
BrowseComp (with context management)accuracy	74.9%thinking · tools: browser, code_interpreter, search · discard-all(setup not fully disclosed)	—
BrowseComp (with tools)accuracy	—	83.2%thinking(setup not fully disclosed)
DeepSearchQAf1	—	92.5%(setup not fully disclosed)
SWE-Bench Proresolved_rate	—	58.6%(setup not fully disclosed)
SWE-Bench Verifiedresolved_ratesetups differ by: harness, reasoning, runs	76.8%non-thinking · moonshot-internal-agent · 5 runs(setup not fully disclosed)	80.2%(setup not fully disclosed)
OSWorld Verifiedsuccess_rate	—	73.1%(setup not fully disclosed)
MMMU Proaccuracy	—	79.4%(setup not fully disclosed)
Humanity's Last Exam Full (with tools)accuracy	50.2%thinking · tools: browser, code_interpreter, search · retain-latest-tool-round(setup not fully disclosed)	—
Humanity's Last Exam Full (with tools)accuracy	—	54%thinking · tools: browser, code_interpreter, search(setup not fully disclosed)
GPQA Diamondaccuracy	—	90.5%thinking(setup not fully disclosed)