What changed
- Gemini 2.5 Flash: $0.10 per 1M input / $0.30 output (was $0.15/$0.45).
- Gemini 2.5 Pro: $0.85 per 1M input / $3.50 output (was $1.25/$5.00).
- Vertex AI batch tier kept its 50% discount.
What it means
For cost-sensitive high-volume workloads (autocomplete, basic classification, RAG synthesis on small contexts), Gemini 2.5 Flash is now the most aggressive price point in the frontier-adjacent tier. Pro is closer to GPT-5-mini / Claude Haiku 4.5.
The production decision tree for cost-tier routing now usually has Flash in the cheap-path slot. Worth running a 1-week eval against your current model on your real workload — quality has been competitive on most extraction / classification / short-form generation tasks.