Best for balancing reasoning and speed. Gemini 2.5 Flash offers thinking capabilities with strong performance across coding, math, and reasoning tasks at an efficient price point.
Gemini 2.5 Flash is Google's generally available 2.5-series workhorse model, stabilized on June 17, 2025 after the 05-20 preview shown at Google I/O. It is built for price-performance: a 1,048,576-token input window, 65,536-token output limit, multimodal input (text, images, video, and audio on the native Gemini API), text output, tool use, structured outputs, code execution, URL context, search grounding, and controllable thinking budgets.
For TheRouter, Gemini 2.5 Flash is the default Google choice when a team needs long-context reasoning and agentic utility without paying Pro-tier prices. It is stronger and more controllable than Gemini 2.0 Flash, less expensive than Gemini 2.5 Pro, and more capable than Flash-Lite for coding, multi-step analysis, retrieval-heavy support, and high-volume document workflows. If raw frontier coding accuracy matters more than cost, route to Gemini 3 Flash/Pro or GPT-5.5-class models instead.
| Type | Rate |
|---|---|
| Input | $0.324 / 1M tokens |
| Output | $2.70 / 1M tokens |
| Release date | 2025-06-17 (stable GA)developers.googleblog.com ↗ | verified |
| Knowledge cutoff | January 2025ai.google.dev ↗ | verified |
| Input / output modalities | Native Gemini API: text, images, video, audio in; text out. TheRouter exposes text/image/PDF input and text output.ai.google.dev ↗ | verified |
| Native tool capabilities | Function calling, structured outputs, code execution, file search, URL context, search grounding, Maps grounding, caching, Batch API, flex and priority inferenceai.google.dev ↗ | verified |
| License | Proprietary API model; available under Google Gemini API / Google Cloud terms, not open weightsai.google.dev ↗ | verified |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
Global-MMLU-Lite Aggregator-reported benchmark for Gemini 2.5 Flash; use as directional until refreshed against Google model-card tables. | 88.4%% | llm-stats.com ↗ | |
AIME 2024 Aggregator-reported math result; Google separately describes the 05-20 Flash update as improved on reasoning benchmarks. | 88.0%% | llm-stats.com ↗ | |
GPQA Aggregator-reported science/reasoning score; keep confidence medium until matched to the underlying model-card methodology. | 82.8%% | llm-stats.com ↗ | |
MMMU Aggregator-reported multimodal score for the stable Flash line. | 79.7%% | llm-stats.com ↗ |
Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.
curl https://api.therouter.ai/v1/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer $THE_ROUTER_API_KEY" -d '{
"model": "google/gemini-2.5-flash",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'Use Gemini 2.5 Flash as an OpenAI-compatible chat model through TheRouter when you want Google long-context reasoning without changing SDKs.
curl https://api.therouter.ai/v1/chat/completions -H "Authorization: Bearer $THEROUTER_API_KEY" -H "Content-Type: application/json" -d '{"model":"google/gemini-2.5-flash","messages":[{"role":"user","content":"Summarize the migration risk in this pull request."}]}'Google made Gemini 2.5 Flash generally available, simplified pricing to a single thinking/non-thinking tier, and positioned Flash-Lite as the cheaper high-throughput option.
Google described the updated Flash preview as better across reasoning, multimodality, code, and long context, while using 20-30% fewer tokens in internal evaluations.
Yes for cost-balanced production traffic. It gives teams 1M context, thinking support, tool calling, structured outputs, and multimodal input at much lower cost than Pro-tier models. For frontier autonomous coding, evaluate Gemini 3 or GPT-5.5-class models instead.
Flash is the stronger price-performance model for reasoning and agentic workflows; Flash-Lite is optimized for the lowest latency and cost, with thinking off by default and a better fit for classification, summarization, and simple extraction at scale.
Yes. TheRouter exposes the model through OpenAI-compatible chat completions with tools in supported_params. Native Gemini tool features beyond generic function calling should be validated against the specific TheRouter route before marketing them as available.
| source | URL | retrieved | |
|---|---|---|---|
| Release date | developers.googleblog.com ↗ | 2026-06-03 | verified |
| Knowledge cutoff | ai.google.dev ↗ | 2026-06-03 | verified |
| Input / output modalities | ai.google.dev ↗ | 2026-06-03 | verified |
| Native tool capabilities | ai.google.dev ↗ | 2026-06-03 | verified |
| License | ai.google.dev ↗ | 2026-06-03 | verified |
| Global-MMLU-Lite | llm-stats.com ↗ | 2026-06-03 | to verify |
| AIME 2024 | llm-stats.com ↗ | 2026-06-03 | to verify |
| GPQA | llm-stats.com ↗ | 2026-06-03 | to verify |
| MMMU | llm-stats.com ↗ | 2026-06-03 | to verify |
| Gemini 2.5 Flash becomes stable GA | developers.googleblog.com ↗ | 2026-06-03 | verified |
| Google I/O update improves Flash efficiency and reasoning | blog.google ↗ | 2026-06-03 | verified |
| Is Gemini 2.5 Flash the right default Google model on TheRouter? | ai.google.dev ↗ | 2026-06-03 | to verify |
| How is Gemini 2.5 Flash different from Gemini 2.5 Flash-Lite? | developers.googleblog.com ↗ | 2026-06-03 | to verify |
| Does Gemini 2.5 Flash support OpenAI-compatible tool calling through TheRouter? | ai.google.dev ↗ | 2026-06-03 | to verify |