Google: Gemini 2.5 Pro Preview 05-06

Researchers and engineers who routinely wrestle long, mixed-format inputs into a single reasoning pass are the natural audience for this dated preview. The 1,048,576-token context window, larger than most models listed here, lets such a user load an entire technical report, its source code, and supporting figures at once, then lean on the built-in chain-of-thought mode to trace a bug or reconcile conflicting findings without splitting the work across prompts. Vision input covers embedded images, and structured JSON output plus function calling make it usable inside tool-driving agents. As an earlier May snapshot of the Pro line, it is aimed at teams pinning a specific build for reproducible evaluation rather than tracking a moving target. On price it matches the Pro tier at $1.25 per million tokens, four times the $0.30 Flash rate, so the spend buys reasoning depth over throughput; against lighter variants like Gemini 3.1 Flash Lite, it retains full reasoning instead of trading it for latency. There is no free tier. It best serves technical users doing content-creation, research, and writing who need a fixed, reproducible endpoint with strong reasoning over long context. Pick it when version stability for testing matters; anyone wanting the latest fixes or lower cost should choose a newer build or the Flash tier instead.