Gemini 2.5 Pro: context specification distinguishes the 1,048,576-token input limit from the 65,536-token output limit
Google DeepMind clarifies the technical ceiling for Gemini 2.5 Pro's massive context window
Context specification distinguishes the 1,048,576-token input limit from the 65,536-token output limit — verified by the OnAirToday AI Progress Tracker against Google DeepMind's published documentation.
OnAirToday
Newsroom

The OnAirToday AI Progress Tracker recorded a verified change to Google DeepMind's Gemini 2.5 Pro (version gemini-2.5-pro): Context specification distinguishes the 1,048,576-token input limit from the 65,536-token output limit.
Gemini 2.5 Pro is Google's multimodal reasoning model for complex analysis, coding and long-context tasks. Its published Gemini API specification supports text, image, audio and video inputs with text output; standard token pricing increases for prompts exceeding 200,000 tokens.
What changed in this review:
- Pricing distinguishes prompts up to 200,000 tokens from higher-priced longer prompts and includes billable thinking tokens.
- Google's technical report documents GPQA Diamond at 86.4% and SWE-bench Verified at 67.2% under its evaluation configurations.
- Published benchmark coverage includes AIME 2025, LiveCodeBench v6, Humanity's Last Exam, MMMU and Aider Polyglot.
- Capability score 56 → 72
- Flaws and limitations updated
- Risks and threats updated
Teams building on Gemini 2.5 Pro should confirm limits, pricing and benchmark coverage against the vendor's own documentation before relying on these figures.
The full profile — capabilities, benchmark citations, known flaws and risk guidance — is on the tracker at /ai-models/gemini-2-5-pro.
References:
- https://ai.google.dev/gemini-api/docs/models#gemini-2.5-pro
- https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-pro
- https://arxiv.org/abs/2507.06261
- https://ai.google.dev/gemini-api/docs/thinking
Generated insight by OnAirToday AI
The massive disparity between input and output limits highlights the industry's focus on processing vast datasets rather than generating long-form content. This distinction is critical for developers building RAG-based systems that require deep analysis of extensive documentation.

Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.