Gemini 3.1 Pro

Google · Released 19 Feb 2026 · gemini-3.1-pro-preview

Compare this model

ProprietaryPreviewReasoningCodingAgentsMultimodal

$2.00

Input price / 1M tokens

#36 of 61, cheapest first

$12.00

Output price / 1M tokens

#44 of 61, cheapest first

1.05M

Context window

#11 of 71

114

Output tokens / second

#26 of 61

44.4%

Humanity’s Last Exam (no tools)

#4 of 23

77.1%

ARC-AGI-2

#3 of 6

94.3%

GPQA Diamond

#3 of 27

68.5%

Terminal-Bench 2.0

#3 of 7

Summary

Gemini 3.1 Pro is a proprietary model from Google, released on 19 Feb 2026.

At $2.00 input and $12.00 output per million tokens, it is #36 of 61 on input price, cheapest first.

Its 1.05M-token context window ranks #11 of 71.

Measured output speed is 114 tokens per second, #26 of 61.

Google's current Pro model, still preview. Built for multimodal understanding, agentic work and vibe-coding, with a focus on software engineering and multi-step tool use.

Benchmarks · 15 reported

Benchmark	Score	Bar	Rank	Source
Humanity’s Last Exam (no tools)no tools, full set text + multimodal, thinking high	44.4%		#4 of 23	Google Gemini 3.1 Pro evaluation reportSelf-reported
Humanity's Last Examsearch (blocklist) + code, thinking high	51.4%		–	Google Gemini 3.1 Pro evaluation reportSelf-reported
ARC-AGI-2ARC Prize Verified, semi-private set	77.1%		#3 of 6	ARC Prize, cited in Google's evaluation report
GPQA Diamondno tools	94.3%		#3 of 27	Google Gemini 3.1 Pro evaluation reportSelf-reported
Terminal-Bench 2.0Terminus 2 harness	68.5%		#3 of 7	Google Gemini 3.1 Pro evaluation reportSelf-reported
SWE-bench Verifiedsingle attempt, average of 10 runs; includes a 0.6-point adjustment for 3 broken harness items	80.6%		#1 of 12	Google Gemini 3.1 Pro evaluation reportSelf-reported
SWE-bench Prosingle attempt, average of 5 runs	54.2%		#14 of 16	Google Gemini 3.1 Pro evaluation reportSelf-reported
LiveCodeBench Propublic leaderboard	2,887	Elo-style	–	LiveCodeBench Pro leaderboard, cited in Google's evaluation report
APEX-Agents	33.5%		–	Google Gemini 3.1 Pro evaluation reportSelf-reported
τ²-benchretail (99.3% telecom)	90.8%		#2 of 6	Google Gemini 3.1 Pro evaluation reportSelf-reported
MCP Atlaspublic set, sourced from Turing	69.2%		–	Turing, cited in Google's evaluation report
BrowseCompDeep Research with search, Python and browsing	85.9%		#3 of 7	Google Gemini 3.1 Pro evaluation reportSelf-reported
MMMU-Prono tools	80.5%		#3 of 12	Google Gemini 3.1 Pro evaluation reportSelf-reported
MMMLU	92.6%		–	Google Gemini 3.1 Pro evaluation reportSelf-reported
MRCR v2 (8-needle)128k average (26.3% at 1M pointwise)	84.9%		#3 of 13	Google Gemini 3.1 Pro evaluation reportSelf-reported