Gemini API Pricing for Developers: How Flash and Pro Costs Actually Work

Gemini API pricing follows a tiered structure: cheap Flash models for high-volume production traffic, Gemini 3.5 Flash for richer agent loops, and separate Pro/Image tiers for specialized workloads. Google Cloud's Agent Platform pricing page lists Gemini 3.5 Flash at $1.50/$9.00 per million input/output tokens on the standard Global tier, while Google's Gemini Developer API pricing page lists Gemini 3.1 Flash-Lite at $0.25/$1.50. If you're choosing between Gemini, GPT-5, and DeepSeek for a specific workload, the decision usually comes down to three factors: token volume, latency tolerance, and reasoning complexity.

Key Takeaways

Source Snapshot

Source	What It Confirms	Observed Date	Link
Google AI Gemini API pricing page	Gemini 3.1 Flash-Lite, Gemini 3.1 Flash Image, Gemini 3 Pro Image, and Veo pricing	2026-07-08	ai.google.dev/gemini-api/docs/pricing
Google Cloud Agent Platform pricing	Gemini 3.5 Flash standard, cached-input, Priority, and Flex/Batch token pricing	2026-07-08	cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing
TokenLab platform pricing data	Per-token rates for GPT-5, Claude, DeepSeek, and TokenLab-routed model comparisons	2026-07-08	Aggregated from provider rate cards; verify against each vendor's own docs

Note: Gemini 3.5 Flash appears on the Google Cloud Agent Platform price sheet, while Gemini 3.1 Flash-Lite and image models appear on the Gemini Developer API price sheet. Do not mix those surfaces without checking which product your workload actually uses.

Model And Cost Comparison

All figures are $ per million tokens (input/output) unless noted otherwise.

Model	Provider	Input $/MTok	Output $/MTok	Best For
gemini-3.1-flash-lite	Google	$0.25	$1.50	High-volume, low-complexity tasks
gemini-3.5-flash	Google	$1.50	$9.00	Balanced production traffic
gemini-3.1-flash-image	Google	$0.50	$3.00	Image-conditioned generation
gemini-3-pro-image	Google	$2.00	$12.00	High-fidelity image generation; image output billed separately
gpt-5-nano	OpenAI	$0.05	$0.40	Ultra-low-cost lightweight tasks
gpt-5-mini	OpenAI	$0.25	$2.00	Cost-sensitive general workloads
gpt-5	OpenAI	$1.25	$10.00	General-purpose production use
gpt-5-pro	OpenAI	$15.00	$120.00	Frontier reasoning, critical accuracy
claude-haiku-4-5	Anthropic	$1.00	$5.00	Fast, low-cost assistant tasks
claude-sonnet-5	Anthropic	$2.00	$10.00	Mid-tier reasoning and code
claude-opus-4-8	Anthropic	$5.00	$25.00	Deep reasoning, agentic workflows
deepseek-v4-flash	DeepSeek	$0.147	$0.294	Cheapest viable production option
deepseek-v4-pro	DeepSeek	$0.441	$0.882	Budget reasoning tasks

For video, TokenLab bills Veo 3.1 as a flat $0.20/second (standard) or $0.08/second (fast), regardless of resolution. Google's own tiered pricing for the same models ranges from $0.05/second (Lite, 720p) to $0.60/second (Standard, 4K with audio) - check which billing model applies to your account before estimating video costs at scale.

// Google Gemini Developer API image model pricing (USD per 1M tokens)
// Source: https://ai.google.dev/gemini-api/docs/pricing (verified 2026-07-08)
const GEMINI_IMAGE_PRICING = {
  "gemini-3.1-flash-image": {
    standard: {
      input: 0.50,
      textOutput: 3.00,
      imageOutput: 60.00,
    },
    batch: {
      input: 0.25,
      textOutput: 1.50,
      imageOutput: 30.00,
    },
  },
  "gemini-3-pro-image": {
    standard: {
      input: 2.00,
      textOutput: 12.00,
      imageOutput: 120.00,
    },
  },
};

Decision Framework: Gemini vs GPT-5 vs DeepSeek

Use this to narrow down a provider for your specific workload rather than guessing:

Your Workload	Recommended Tier	Why
High-volume simple tasks (tagging, routing, moderation)	gemini-3.1-flash-lite or deepseek-v4-flash	Lowest cost per token; simple tasks don't need frontier reasoning
Chat/assistant with moderate complexity	gemini-3.5-flash or gpt-5-mini	Similar price-to-quality ratio, both handle multi-turn context well
Code generation or agentic workflows	gpt-5 or claude-sonnet-5	Stronger tool-use and multi-step reasoning benchmarks at this price point
Long-context document analysis	gemini-3.5-flash or deepseek-v4-pro	Start with current high-context models, then benchmark against task-specific needs
Mission-critical accuracy (legal, financial, medical review)	gpt-5-pro or claude-opus-4-8	Highest reasoning ceiling justifies the premium
Cost-constrained prototyping	deepseek-v4-flash	Cheapest per-token rate in this comparison by a wide margin

If you're still unsure, run a 48-hour A/B test: send the same 500-request sample through your top two candidates, log actual token usage, and compare real cost-per-task rather than list price. Token counts vary by model tokenizer, so list price alone can be misleading.

When To Use TokenLab

1. Unified billing across providers. Instead of managing separate accounts and invoices for Google, OpenAI, Anthropic, and DeepSeek, TokenLab consolidates usage into one dashboard so you can compare actual spend, not just list prices.
2. Real-time cost tracking per model. When testing Gemini Flash against GPT-5-mini or DeepSeek, TokenLab shows token consumption and cost side by side, making the A/B test in the decision framework above straightforward to run.
3. Simplified video billing. If Veo 3.1's tiered resolution pricing is hard to forecast, TokenLab's flat per-second rate removes that variable from your budget planning.

Related Reading

• AI API pricing comparison
• TokenLab migration guide

FAQ

How do I know whether to pick Gemini, GPT-5, or DeepSeek for my workload? Match your task complexity to the decision framework table above. Simple, high-volume tasks favor Gemini Flash-Lite or DeepSeek Flash on price alone. Complex reasoning or agentic tasks favor GPT-5 or Claude Opus, where the higher per-token cost is usually offset by fewer failed completions and retries. When in doubt, run a small real-workload test before committing to a provider.

Are these prices guaranteed to be current? No. Gemini 3.1 Flash-Lite and Gemini image pricing are tied to Google's Gemini Developer API pricing page, while Gemini 3.5 Flash is tied to Google Cloud Agent Platform pricing. GPT, Claude, and DeepSeek rows reflect TokenLab's tracked platform pricing as of the same date. Because providers update rate cards without much warning, verify exact numbers on the provider's official pricing page before finalizing a budget or invoice estimate.

Why isn't there a GPT-5 comparison with speculative future models? This article only uses pricing for models that are actually listed in TokenLab's current catalog. Estimated or unreleased model pricing isn't included, since unverified figures make cost planning unreliable - a core reason to prefer sourced, dated pricing data over projections.

Does Gemini 3.5 Flash cost more than GPT-5-mini for the same task? Yes, on a per-token basis: gemini-3.5-flash is $1.50/$9.00 per MTok versus gpt-5-mini at $0.25/$2.00 per MTok. However, actual cost depends on tokens consumed per task, which varies by model. Test both on your real prompts before assuming one is cheaper overall.

Is DeepSeek always the cheapest choice? On list price, yes - deepseek-v4-flash at $0.147/$0.294 per MTok is the lowest in this comparison. But cost-per-task also depends on output verbosity and retry rates, so validate quality against your specific use case before switching a production workload.

Ready to compare Gemini, GPT-5, Claude, and DeepSeek costs on your actual workload instead of guessing from list prices? Start tracking real usage costs with TokenLab.