Gemini 2.5 Pro

Gemini 2.5 Pro arrived on 25 March 2025 as an experimental release and became the model that defined Google's 2025. Its two defining traits were a context window of roughly a million tokens — inherited from Gemini 1.5 Pro but now paired with frontier-level reasoning — and thinking that was switched on by default rather than sold as a separate product line. Where OpenAI split its catalogue into ordinary models and reasoning models, Google folded the two together and let developers set how much thinking they were willing to pay for. Google reported strong figures at launch: 84.0% on GPQA Diamond, 86.7% on the AIME 2025 mathematics qualifier at a single attempt, 18.8% on Humanity's Last Exam without tools, and 63.8% on SWE-bench Verified with a custom agent harness. The model went straight to the top of the LMArena leaderboard, a position Google had been chasing since the first Gemini. General availability followed on 17 June 2025, with pricing that undercut most rivals at the frontier: 1.25 dollars per million input tokens for prompts up to 200,000 tokens and 2.50 dollars above that, with output at 10 and 15 dollars. That combination — long context, native multimodality across text, images, audio and video, and a price well below the competition — made it the default choice for document-heavy and codebase-wide work for much of the year. A higher-compute variant, Deep Think, reached Google's top-tier subscribers in August 2025 and was the configuration Google used for its most demanding reasoning demonstrations. The line has since been succeeded by Gemini 3 Pro and Gemini 3.1 Pro. On Vertex AI, Google lists 16 October 2026 as the earliest date on which gemini-2.5-pro may be retired; the developer-facing Gemini API documentation still shows no announced shutdown date, and Google has promised notice in advance.

▸Videos

No videos yet.

▸Reviews