GEMINI 2.5 PRO
1M-token context. Multimodal. Strong reasoning across code, math, and STEM. Native googleSearch grounding. $1.25/$10 per million tokens.
Gemini 2.5 Pro is Google's previous-generation flagship and still a serious frontier model in mid-2026. It's a thinking model — extended chain-of-thought reasoning before the final answer — with strong scores across code, math, and STEM. The 1M-token native context window remains industry-leading among non-beta models, and the full multimodal stack (text/image/audio/video) is intact. Pricing at $1.25/$10 sits below Gemini 3 Pro ($2.50/$10) but above Flash 3 ($0.50/$3). The case for: long-context analysis of large datasets and codebases, math/STEM reasoning, anywhere you want Gemini 3 Pro quality on a tighter budget. The case against: most new work should default to Gemini 3 Pro — it's a clear step up in coding, tool use, and reasoning at only 2x the input cost. Keep 2.5 Pro for workloads already validated against it.
For new work, default to Gemini 3 Pro — it's a meaningful upgrade across coding, reasoning, and tool use, and the price gap is only 2x on input. Keep 2.5 Pro live for workloads already validated against it where the output is known-good, and for budget-sensitive long-context jobs where the slightly cheaper input price ($1.25 vs $2.50) matters at volume.
Yes — it produces an extended chain-of-thought before the final answer. The reasoning is visible in the API response and improves performance on hard math, code, and multi-step problems.
Gemini 3 Pro is a clear upgrade — better coding, better tool use, sharper reasoning, slightly better long-context coherence. Price gap is 2x on input ($1.25 vs $2.50) and equal on output ($10).
$1.25 per million input tokens and $10.00 per million output tokens. Council AI bundles it inside monthly plan budgets.
Yes — it remains an approved model in the lineup for users on Pro and Ultra plans. New chats default to Gemini 3 Pro but 2.5 Pro is selectable.
Long-context analytical work — whole-codebase code review, document analysis, dataset summarization, STEM problem-solving — where the thinking-mode output is already validated for your task.