News - Jul 3, 2026
Gemini 2.5 Pro Gets Smarter, But Free Tier Gets Squeezed
Google has updated Gemini 2.5 Pro, its flagship reasoning model, with measurable benchmark improvements across coding, math, and multimodal tasks. The model now tops several key leaderboards including LMArena, where it outscores competitors on human preference ratings. Specifically, Google reports improved performance on GPQA Diamond (graduate-level science questions) and stronger scores on SWE-bench, a coding evaluation that tests real software engineering tasks. On the access side, Gemini 2.5 Pro remains available through Google AI Studio for free at rate-limited capacity, and through the Gemini API at $1.25 per million input tokens (under 200k context) and $10 per million output tokens. Google One AI Premium subscribers at $19.99 per month get priority access through the Gemini app, but the free tier now faces tighter daily usage caps than at launch. Who benefits most: developers building coding assistants or complex reasoning pipelines, and researchers who need long-context handling up to one million tokens. Who gets squeezed: casual free users who were leaning on the model daily will hit walls faster. Enterprise teams on tight budgets will also feel the output token pricing, which remains steep compared to alternatives like Claude 3.5 Sonnet or GPT-4o.
What changed
Improved benchmark scores on GPQA Diamond and SWE-bench; tighter free-tier usage caps; pricing holds at $1.25 per million input tokens and $10 per million output tokens; available via AI Studio, Gemini API, and Google One AI Premium at $19.99 per month.
Who this affects
Developers and researchers using coding, math, or long-context tasks benefit most. Free users and budget-conscious enterprise teams get the short end.
Our take
Better benchmark numbers mean little if the pricing structure pushes the people who actually need this model toward cheaper, nearly-as-good alternatives.
Based on Google DeepMind announcements and publicly available API pricing as of mid-2025. No affiliate relationship with Google.
VerdictBloom is editorially independent. No company reviewed or approved this article before publication.