Google Gemini API Errors Directory — Fixes for Gemini 2.5 Pro, Flash, Vertex AI (2026)
Home › Providers › Gemini

Google Gemini API Errors — Complete Fix Directory

Every Gemini error, decoded and fixed. Gemini 2.5 Pro, 2.5 Flash, thinking mode, context caching, safety filters, Vertex AI vs AI Studio. Tested against the live API, November 2026.

15
Error pages live
7
Categories covered
2
Deployment paths
Nov 2026
Last tested
gemini-2.5-pro gemini-2.5-flash gemini-2.5-flash-lite gemini-2.0-flash gemini-embedding-001 text-embedding-004

Two deployment paths, same models

Google AI Studio (generativelanguage.googleapis.com): API-key based, fast prototyping, free tier available. Simpler auth, works from anywhere.

Vertex AI (aiplatform.googleapis.com): Google Cloud IAM, enterprise controls, regional endpoints, VPC-SC support. Requires GCP project + service account.

Many errors are path-specific — check which you're using before searching for a fix.

Q

Quota & Authentication

Rate limits, model availability, API key issues
429 RESOURCE_EXHAUSTED
Quota Exceeded (429)
RPM, TPM, RPD limits. Free vs paid tier quotas, backoff strategies, quota project mapping.
 404 NOT_FOUND
Model Not Available in Region
Vertex AI regional availability. Model rollout gaps, endpoint routing.
 403 PERMISSION_DENIED
API Key Invalid or Restricted
Referrer/IP restrictions, expired keys, API not enabled, AI Studio vs Vertex confusion.
F

Function Calling & Controlled Generation

Tool schemas, response_schema, JSON mode
400 INVALID_ARGUMENT
Function Call Schema Invalid
OpenAPI 3.0 subset, unsupported types, format restrictions. Migration from OpenAI schemas.
 400 response_schema
Controlled Generation Schema Error
response_mime_type + response_schema pattern. Enum, nullable, propertyOrdering quirks.
S

Safety Filters

Content policy, harm categories, threshold tuning
finish_reason SAFETY
Safety Filter Blocked Response
SAFETY finish_reason, empty candidates, prompt_feedback.block_reason. Handling gracefully.
 safety_settings
Harm Category Threshold Config
HARM_CATEGORY_* configuration, thresholds, false positives, civic integrity category.
T

Thinking Mode (Gemini 2.5)

thinking_budget, reasoning tokens, model support
thinking_budget
Thinking Budget Exceeded
Empty response when budget too tight. 2.5 Pro vs Flash budgets, dynamic thinking.
 400 field not supported
Thinking Not Supported on Model
thinking_config only on 2.5 series. Migrating from Flash-Lite or 1.5. Cross-model helpers.
C

Context & Caching

Long context, cached content, token counting
400 context length
Context Window Exceeded
2M tokens on 2.5 Pro, 1M on Flash. Counting with countTokens, video+audio tokens.
 cache miss / TTL
Context Caching Not Hitting
Cached content TTL, min 4096 tokens, model+content match. Discount not applied.
M

Multimodal Inputs

Video, image, audio processing
400 token overflow
Video Token Limit Exceeded
Video ~300 tokens/sec at default fps. Trimming, mediaResolution, sampling strategies.
 400 unsupported MIME
Image MIME Type Unsupported
Only PNG, JPEG, WebP, HEIC, HEIF. Converting BMP, GIF, TIFF. inline vs File API.
V

Vertex AI vs AI Studio

Auth, endpoints, feature parity gaps
401 UNAUTHENTICATED
Vertex AI Auth Failed
Service account credentials, ADC, scopes, IAM role misconfigurations, token refresh.
 quota mismatch
AI Studio vs Vertex Quota Confusion
Different quota pools per deployment path. Free tier only on AI Studio, GCP billing on Vertex.