Loading...
Google · Anthropic — Interactive pricing comparison
Gemini 2.5 Pro edges ahead on raw coding and reasoning scores, but Sonnet 4 is the only one here that can drive a computer. Same 1M context on both sides means the real decision is whether you need agentic control or just stronger benchmarks.
Anthropic
Gemini 2.5 Pro is 40% cheaper at this usage
Gemini 2.5 Pro has a larger context window
The headline number lies a little. Gemini's $1.25 input price looks like a clear win against Sonnet's $3, and on input-heavy work—long document analysis, large-context retrieval, summarization—it genuinely is. But flip to output-heavy work and the gap collapses: Gemini bills $10 per million output tokens against Sonnet's $15, a difference that mostly disappears once your pipeline starts generating long completions, agent traces, or multi-step code. So the cost story isn't 'Gemini is cheaper.' It's 'Gemini is cheaper when you're feeding it more than it writes back.' For a codegen agent that emits thousands of lines, you're paying near-Sonnet rates anyway—at which point Sonnet's computer-use capability is the deciding factor you didn't pay extra for. Benchmark-wise the two trade single digits (Gemini 60/65 vs Sonnet 58/64), close enough that neither wins on capability alone. The cleaner mental model: price your actual token ratio first, then ask whether computer-use is on your roadmap. Teams that skip the first step and anchor on the input price tend to be surprised by their output-heavy bill.
Last updated June 2026