Gemini 3 Pro

Gemini 3 Pro is Google DeepMind's late-2025 flagship, succeeding Gemini 2.5 Pro with a 1 million token input context window (64K output) and native text/image/audio/video understanding. It is API-accessible through Google AI Studio and Vertex AI, and powers the Gemini app's paid tiers.

At launch it broke 1500 Elo on LMArena, reaching 1501 and the #1 public ranking — genuine independent human-preference validation. Google also reported best-in-class scores on graduate-level science (GPQA Diamond 91.9%), abstract reasoning (ARC-AGI-2 45.1% via Deep Think, an extended-reasoning mode), and multimodal understanding (MMMU-Pro).

It introduced a tiered pricing model: $2/$12 per 1M input/output tokens for prompts up to 200K tokens, doubling beyond that. Google has since moved its flagship spotlight to Gemini 4 Argon, but Gemini 3 Pro remains widely used as a long-context reasoning workhorse.

Provider-reported (Google DeepMind):

• GPQA Diamond: 91.9% (with Deep Think)
• ARC-AGI-2: 45.1% (with Deep Think — ~2x Gemini 2 Pro)
• MMMU-Pro: 81% (multimodal reasoning)
• "19 out of 20" internal benchmark comparisons vs. rivals

Independent (LMArena): 1501 Elo at launch — first model to break 1500, #1 public ranking on release. SWE-bench: Google highlighted 78% for the Gemini 3 Flash variant, not Pro — Pro's coding score was not prominently disclosed (noted as a transparency gap by third-party reviewers).