LLM & ChatActive
google / gemini-3.7-flash
gemini-3.7-flash
Google’s Gemini 3.7 Flash is a multimodal model built for coding and agent workflows. It takes text plus images, audio, or video and returns text.
Partner LLM
Model ID
gemini-3.7-flash
Provider
google
Updated
1787060996
wiro playground—google/gemini-3.7-flash
Updated 1787060996
Overview
Gemini 3.7 Flash is a Gemini 3 family model from Google, tuned for coding and agent workflows. You provide text, plus optional images, audio, or video for context. The model applies an adjustable “thinking” mode to plan and reason through multi-step work. It returns text outputs up to 64K tokens, which helps when you need full code patches, long reports, or detailed plans.
What you can build
- Code review and debugging from a large codebase pasted into one session
- Agent task plans that break work into steps and checkpoints
- Web app scaffolds, UI copy, and iterative front-end fixes from screenshots
- Document analysis and knowledge-work writeups (finance, legal, bioscience summaries)
- Video Q&A and moment finding when you need answers tied to what appears on screen
- Audio Q&A, light transcription, and timeline-based analysis using time ranges
Inputs
- Your prompt as plain text. This is required.
- Optional images for visual context. Use common formats like PNG, JPEG, WEBP, HEIC, or HEIF.
- Optional audio clips for speech and sound context. Supported formats include WAV, MP3, AIFF, AAC, OGG Vorbis, and FLAC.
- Optional videos for scene understanding. Common supported formats include MP4, MPEG, QuickTime, AVI, WebM, WMV, and 3GPP.
- Optional system instructions as text. Use this to set role, tone, and hard rules.
- Optional thinking depth selection. Gemini 3.7 Flash supports low, medium, and high thinking.
- Optional maximum response length. The model supports up to 65,536 output tokens.
- Optional user identifier (string or number). Provide this to keep chat history tied to a person.
- Optional session identifier (string or number). Provide this to keep a multi-turn thread consistent.
Outputs
- Text output only.
- Long-form answers with structure like headings, numbered steps, tables, and code blocks.
- Multi-step agent-style plans that include what to do next and what to verify.
- Responses can be very long when you allow a large output token limit.
Recommended settings
- Use medium thinking for most coding and agent tasks. This is the default for Gemini 3.7 Flash.
- Use high thinking for long-horizon debugging, issue resolution, or tasks needing careful planning.
- Use low thinking for short Q&A, fast classification, or simple rewrites.
- Cap the output token limit when you only need a short answer. Raise it for full patches and reports.
Limitations
- The model can hallucinate. Verify facts, citations, and exact API details.
- The knowledge cutoff is March 2026. Some areas may behave like January 2025 coverage.
- The model is still improving jailbreak resistance. Don’t rely on it for policy bypass detection.
- You may see occasional slowness or timeouts during high demand.
- Text is the only output modality. It does not generate images, audio, or video.
- Low-quality inputs reduce accuracy. Blurry screenshots, noisy audio, and low-resolution video can mislead it.
- Very long or cluttered documents can cause missed details. Break work into smaller chunks.
Safety & compliance
- Google ships Gemini 3.7 Flash with updated safeguards against misuse in CBRN and cyber offense domains.
- Follow Google’s acceptable use rules for Gemini models.
- Don’t submit sensitive personal data unless you have a policy and user consent to do so.
- Treat outputs as suggestions, not professional advice for legal, medical, or safety-critical decisions.
API quick start
Run gemini-3.7-flash with a single API call.
POST https://api.wiro.ai/v1/Run/google/gemini-3.7-flash
{
"prompt": "Tell me about Türkiye",
"inputAll": "https://your-cdn.com/input.png",
"user_id": "...",
"session_id": "..."
}