Google · updated 2026-08-29
Run Gemini 2.5 (Pro / Flash): providers, pricing and flat-rate access
Google's 2.5 generation — Pro for reasoning, Flash for speed — long a workhorse for multimodal and long-context work, now late in its lifecycle with Gemini 3.x as Google's recommended line.
Gemini 2.5 runs via the Gemini API per-token (Pro $1.25/$10 per M tokens, Flash $0.30/$2.50; Pro is paid-only since April 2026), via Google AI Pro/Ultra plans ($19.99–199.99/mo, request-per-day caps in Gemini CLI), or inside Standard Compute's flat-rate routed pool, where both Pro and Flash serve requests at a fixed monthly price.
Your options, honestly
Google AI Pro / Ultra$19.99–199.99/moApp-and-CLI plans with daily request caps (1,500–2,000/day in Gemini CLI) — predictable, but capped and Google-only.
Gemini API (AI Studio)Pay-per-tokenPro $1.25/$10, Flash $0.30/$2.50 per M tokens; prompts over 200K tokens bill higher (Pro $2.50/$15). Free tier no longer covers Pro.
Vertex AIPay-per-tokenEnterprise route with cloud billing and quotas; same per-token economics.
Standard Compute$39+/mo flat2.5 Pro and Flash in the routed pool with other frontier models — no daily caps, one price. Routing decides when Gemini-class is used; no pinning.
the flat-rate trade-off
The honest caveat first: Google has scheduled Gemini 2.5 for retirement in October 2026 and points new work at 3.x — so don't build a 2.5-only dependency anywhere. In a routed pool that risk is Standard Compute's problem, not yours: when a model retires, routing shifts to its successors and your endpoint doesn't change.
Running it in your agent
OpenAI-compatible agents reach Gemini 2.5 through OpenRouter-style gateways or Standard Compute's routed 'standardcompute' model. Google's own route is Gemini CLI under an AI Pro/Ultra plan, or API keys from AI Studio / Vertex.
FAQ
Is it safe to build on Gemini 2.5 now?
Directly, only short-term: Google has slated the 2.5 models for retirement on October 16, 2026. If you use 2.5 via a routed multi-model plan, the retirement is absorbed by routing; if you pin it via API, plan a 3.x migration.
Gemini 2.5 Pro or Flash for agents?
Flash for routine, high-volume steps ($0.30/$2.50 is genuinely cheap); Pro where output quality matters — noting Pro's $10/M output rate is where agent bills grow. Routed plans make that split automatic.
More: how flat-rate LLM APIs work · per-agent guides for OpenCode, Claude Code and Cline · model comparisons