Gemini 2.5 Flash Image
Image AvailableCompare models
ComparingGemini 2.5 Flash Imagewith
Pick a model above to see the comparison.
google/gemini-2-5-flash-image · by Google DeepMind · Diffusion transformer (DiT)
Pricing — 1 offering(s)
Standard · up to 1024x1024 output
- $0.039 / image Current 2026-07-23 → present
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
Capability profile
Text rendering unknown
Inference speed strong
Resolution ceiling weak
Prompt adherence unknown
Photorealism unknown
Artistic style range unknown
Compositional accuracy unknown
Benchmarks
| Benchmark | Score | Config | Source |
|---|---|---|---|
| GenEval | 0.97 | Auto-extracted by benchmark-tracker from arXiv: PromptRL: Prompt Matters in RL for Flow-Based Image Generation. Verify the exact model variant and eval config before merge. | paper ↗ |
| PickScore | 0.97 | Auto-extracted by benchmark-tracker from arXiv: PromptRL: Prompt Matters in RL for Flow-Based Image Generation. Verify the exact model variant and eval config before merge. | paper ↗ |
Operator guidance
Google's own docs now steer new integrations to Nano Banana 2 / Nano Banana 2 Lite (Gemini 3.1 Flash Image family) for better quality, speed, and pricing than this model. Route here only for existing integrations already pinned to gemini-2.5-flash-image, or when its lower Priority-tier cost floor specifically matters; step up to Gemini 3.1 Flash Image (gemini-3-1-flash-image.yaml) for a directly comparable but newer- generation Flash-tier model in the same family.
Use cases
- High-volume, low-latency generation and conversational image editing where cost matters more than peak quality
- Legacy integrations already built against gemini-2.5-flash-image — Google's own docs frame this as still-stable but no longer the recommended entry point
Limitations
- No independently published GenEval / T2I-CompBench score found — this entry's capability ratings lean 'unknown' rather than vendor-repeating marketing claims not present in this model's own current docs
- Single 1024x1024 output resolution only — no higher-resolution tier
- Google's own documentation actively recommends migrating away from this model, calling it 'the legacy pioneer of the Nano Banana series'