# Gemini models — catalogue (Gemini Developer API)

Status: DOCUMENTED + LIVE_DISCOVERED (58 ids via paginated GET /v1beta/models, 22 via GET /v1/models) + LIVE_VERIFIED for the ids probed on 2026-09-19 (see verification per record). Machine-readable twin: generated/fragments/models/gemini-models.json (97 records: 58 live + 39 documented-only/retired). This page is generated by scripts/gen_gemini_models_fragment.py — edit the script, not this file.

Sources: https://ai.google.dev/gemini-api/docs/models · model pages …/docs/models/<id> · https://ai.google.dev/api/models · https://ai.google.dev/gemini-api/docs/pricing · https://ai.google.dev/gemini-api/docs/deprecations · https://ai.google.dev/gemini-api/docs/rate-limits · https://ai.google.dev/gemini-api/docs/thinking · https://ai.google.dev/gemini-api/docs/api-versions · discovery document revision 20260918.

Last verified: 2026-09-18 (docs) / 2026-09-19 UTC (live). Related: pricing · rate limits · deprecations & changelog · auth / headers / versions · errors · SDKs · OpenAI compatibility · Gemini index

# 1. Families and naming

Naming patterns (models.md, convention since Sept 2025): stable gemini-3.6-flash; preview gemini-3.1-pro-preview / dated …-preview-09-2025 (billing on, tighter limits, ≥2 weeks deprecation notice); latest aliases gemini-flash-latest, gemini-pro-latest, gemini-flash-lite-latest, gemini-2.5-flash-native-audio-latest (hot-swapped, 2-week e-mail notice for breaking changes); experimental lyria-realtime-exp (no production use). Live version strings are inconsistent: dated (3.7-flash-08-2026), bare (3.0 for gemini-3.8-flash and all image models), or the display name (Gemini Flash Latest).

# Alias resolution observed

gemini-3-pro-preview was shut down 2026-03-09 and per the changelog 'now points to gemini-3.1-pro-preview'; live GET still returns version: 3-pro-preview-11-2025. nano-banana-pro-preview is the same live object as gemini-3-pro-image-preview (version 3.0, display name Nano Banana Pro). gemini-3.1-pro-preview-customtools is a variant endpoint of gemini-3.1-pro-preview (same version 3.1-pro-preview-01-2026) tuned to prefer custom tools.

# 2. Live catalogue (58 ids) — limits, lifecycle, verification

Shut-down ids still returned by the live listing: gemini-3-pro-preview, gemini-3.1-flash-lite-preview, gemini-3.1-flash-image-preview, gemini-3-pro-image-preview/nano-banana-pro-preview (deprecations.md dates in the past). Conversely imagen-4.0-* (shut down 2026-08-17) is gone: GET → 404 Model is not found … for api version v1beta.

# 3. Model × Capability matrix (live ids)

Legend: Y = supported (docs and/or live), Y* = supported with qualifier (see JSON, e.g. Supported (Preview)), - = not supported, ? = unknown/not documented. Columns: Think=thinking, Struct=structured output (responseSchema/responseJsonSchema), FC=function calling, Search/Maps=grounding, URL=url_context, Code=code execution, CU=computer use, FS=file search, Cache=explicit context caching (live createCachedContent), Batch=batchGenerateContent/asyncBatchEmbedContent, Flex/Prio=service tiers, Live=bidiGenerateContent, Inter.=Interactions API, OAI=OpenAI-compatible chat.

Thinking levels (thinking.md): gemini-3.8-flash default medium, levels low/medium/high; gemini-3.7-flash default medium, levels low/medium/high; gemini-3.6-flash default medium, levels minimal/low/medium/high; gemini-3.5-flash-lite default minimal, levels minimal/low/medium/high; gemini-3.1-pro-preview default high, levels low/medium/high; gemini-3.1-pro-preview-customtools default high, levels low/medium/high; gemini-3.1-flash-lite-image default minimal, levels minimal/high; gemini-3-flash-preview default high, levels minimal/low/medium/high; gemini-3-pro-preview default high, levels low/high; gemini-3.5-flash default medium, levels minimal/low/medium/high; gemini-3.1-flash-lite default unknown (docs table omits it; openai.md maps minimal/low/medium/high), levels minimal/low/medium/high. Gemini 2.5: thinking_budget (2.5 Flash-Lite off by default; 2.5 Pro cannot be disabled). Gemini 3.x: thinking_level replaces thinking_budget (400 if both are sent); minimal is rejected by 3.7/3.8 Flash; thought signatures must be echoed in multi-turn function calling (MISSING_THOUGHT_SIGNATURE).

# 4. Model × Endpoint matrix (supportedGenerationMethods, verbatim from live)

streamGenerateContent is never listed but is implied by generateContent. interactions (POST /v1beta/interactions) is not reported by the Models API; see column Inter. above.

# 5. Model × Tool matrix (built-in tools; docs support tables)

Tool pricing: Google Search $14/1k requests after 5,000 free/month (Gemini 3; $35/1k grounded prompts after 1,500 RPD on 2.5); Maps $14/1k queries (Gemini 3) or $25/1k prompts (2.5); code execution / URL context / computer use billed as tokens; File Search indexing $0.15/1M tokens. Built-in tools can be combined with function calling since 2026-03-18 (tool-combination.md).

# 6. Live vs docs discrepancies

Global: api-versions.md says all models exist in both v1 and v1beta, but GET /v1/models lists only 22 GA ids and GET /v1/models/gemini-3.1-pro-preview → 404. The Gemini 2.5 family returns 404 no longer available to new users on generation for new keys (undocumented). Our key is Free tier: Pro models answer 429 limit: 0.

# 7. Documented-only / retired ids (not in the live listing)

# 8. Preview / availability policy

• Preview models: production use allowed, billing enabled, more restrictive rate limits, ≥2 weeks deprecation notice; -latest aliases: 2-week e-mail notice before a breaking swap.
• Free tier: many text models are free of charge (content used to improve Google products — Unpaid Services terms); Pro, image, video, music, Omni and Deep Research/Antigravity agents need a paid key (observed limit: 0 on Pro).
• Regions: ~190 countries/territories (available-regions.md). EEA/UK/CH: free and paid tiers available to developers, but API clients serving end users there must use Paid Services (terms §Use Restrictions).
• Gemini Enterprise Agent Platform (Vertex AI) offers most Gemini/Veo ids with regional endpoints and enterprise controls — see vertex-vs-gemini-api.

# 9. Live probe summary (2026-09-19)

Estimated cost of all probes: < $0.001 (a few hundred tokens on Free-tier models).