TestsBenchmarks · Long context

MRCR v2 (8-needle, 1M pointwise)

MRCR v2 8-needle at 1M tokens.

% solvedHigher is betterThe test's websiteOfficial page ↗

The best 5 · 5 models tested

Top 5 · best setting per model · 5 models, 5 results (0 independent)

Measured by independent testersIndependent Reported by the makerVendor-reported

MRCR v2 (8-needle, 1M pointwise) leaderboard
				Thinking levelSetting
1	Gemini 3.6 Flash	Google (Gemini / DeepMind)	54.0%	Defaultdefault settings (API default thinking_level=medium)	Maker's own figureVendor-reportedGoogle ↗	—	21 Jul 2026
2	Gemini 3.5 Flash	Google (Gemini / DeepMind)	26.6%	Defaultdefault settings (API default thinking_level=medium)	Maker's own figureVendor-reportedGoogle ↗	—	19 May 2026
3	Gemini 3.1 Pro (Preview)	Google (Gemini / DeepMind)	26.3%	HighThinking (High)	Maker's own figureVendor-reportedGoogle ↗	—	19 Feb 2026
4	Gemini 3.5 Flash-Lite	Google (Gemini / DeepMind)	21.3%	Highhigh thinking	Maker's own figureVendor-reportedGoogle ↗	—	21 Jul 2026
5	Gemini 3.1 Flash-Lite	Google (Gemini / DeepMind)	12.3%	High	Maker's own figureVendor-reportedGoogle ↗	—	3 Mar 2026