Google Gemini API Pricing Guide (2026): Gemini 4 Argon, 3.1 Pro & 3.8 Flash Costs

Google's Gemini API ecosystem represents one of the most aggressive and architecturally unique pricing models in generative AI. Powered by Google's custom TPU (Tensor Processing Unit) infrastructure, Gemini models combine industry-leading context windows (1 Million to 2+ Million tokens) with native multimodal reasoning.

On September 30, 2026, Google expanded this ecosystem with the announcement of Gemini 4 Argon—a frontier model featuring an unprecedented 1 Million token output limit engineered for autonomous software engineering and defensive cybersecurity.

Below is the verified, up-to-date pricing structure across Google's complete 2026 Gemini API lineup.

---

2026 Gemini API Quick Price Matrix

| Model Tier | Input (per 1M) | Output (per 1M) | Cached Input (per 1M) | Max Output Limit | Context Window | | :--- | :--- | :--- | :--- | :--- | :--- | | Gemini 4 Argon (New) | $2 | $10 | $0.1 | 1,000,000 tokens | 2M tokens | | Gemini 3.1 Pro | $2 | $12 | $0.5 | 8,192 tokens | 1M tokens | | Gemini 3.8 Flash | $0.75 | $3.75 | $0.075 | 8,192 tokens | 1M tokens | | Gemini 3.1 Flash-Lite | $0.25 | $1.5 | $0.0625 | 8,192 tokens | 1M tokens |

Note: All Google API models qualify for a 50% discount when using asynchronous Batch Requests.

---

1. Gemini 4 Argon: The 1M Output Frontier Model (New)

Announced on September 30, 2026, Gemini 4 Argon introduces a fundamental architectural breakthrough: expanding generation limits from standard 8K–64K output windows up to 1,000,000 continuous output tokens.

Argon is designed for sustained, multi-day reasoning tasks:

• Autonomous Software Engineering: Refactoring entire monorepos, performing automated legacy framework migrations, and writing end-to-end test suites in a single generation pass.
• Defensive Cybersecurity: Deep-inspecting binary traces, enterprise network logs, and multi-cloud infrastructure configurations.
• Enterprise Knowledge Work: Analyzing multi-hundred-page regulatory filings, financial ledgers, and legal discovery archives.

Pricing Structure:

• Introductory Input Rate: $2 per 1 Million tokens.
• Introductory Output Rate: $10 per 1 Million tokens.
• Prompt Caching Discount (95% off): $0.1 per 1 Million cached tokens.
• Introductory promotional rates are active through November 30, 2026. Standard pricing will transition to $4.00 input / $20.00 output ($0.20 cached) starting December 1, 2026.

---

2. Gemini 3.1 Pro: Tiered Reasoning & Video Intelligence

Gemini 3.1 Pro serves as Google's production workhorse for high-depth multimodal reasoning, competing directly with GPT-5.5 and Claude Opus. It features native, frame-by-frame video understanding and complex audio processing.

Pricing Tiers:

• Standard Context (≤ 200,000 tokens):
  • Input: $2 per 1M tokens
  • Output: $12 per 1M tokens
  • Cached Input: $0.5 per 1M tokens
• Extended Context (> 200,000 tokens):
  • Input: $4.00 per 1M tokens
  • Output: $18.00 per 1M tokens
  • Cached Input: $1.00 per 1M tokens

Because Gemini 3.1 Pro prices base input at only $2/1M, it offers one of the lowest entry barriers for long-document research and video transcripts.

---

3. Gemini 3.8 Flash & Flash-Lite: High-Speed Workhorses

For latency-critical applications, customer-facing chatbots, and high-throughput agent loops, Google offers the Flash family.

Gemini 3.8 Flash:

Shipped in September 2026 with introductory pricing through December 31, 2026:

• Input Rate: $0.75 per 1M tokens.
• Output Rate: $3.75 per 1M tokens.
• Prompt Caching: $0.075 per 1M tokens.
• Standard rates double on January 1, 2027 to $1.50 in / $7.50 out.

Gemini 3.1 Flash-Lite:

Google's ultra-budget tier for high-volume extraction, classification, and embeddings:

• Input Rate: $0.25 per 1M tokens.
• Output Rate: $1.5 per 1M tokens.
• Cached Input: $0.0625 per 1M tokens.

---

4. How Google Billed Multimodal Tokens (Audio, Video & Images)

Unlike competitors who bill static dollar surcharges per image or audio minute, Google standardizes everything into input tokens:

1. Images: Fixed at approximately 258 tokens per image tile (or based on resolution tiles). At Flash rates, processing an image costs roughly $0.00019 (a fraction of a cent).
2. Audio: Billed at roughly 32 tokens per second (or ~1,920 tokens per audio minute).
3. Video: Sampled at 1 frame per second (fps). Each second of 1 fps video equates to 258 image tokens plus audio track tokens.

---

5. Cost-Saving Strategies for Gemini API

• Leverage Prompt Caching: When repeatedly querying documents or system prompts over 32k tokens, caching reduces input costs by 90% to 95%.
• Use the Batch API: If your processing is non-real-time (e.g., nightly scraping, bulk evaluations, translation), Google offers a flat 50% discount across all Gemini models.
• Route High-Volume Work to Flash-Lite: Use Gemini 3.1 Flash-Lite for intent classification and entity extraction, escalating only complex code synthesis to Gemini 4 Argon or Gemini 3.1 Pro.

---

Calculate Your Exact Monthly Costs

Want to simulate your workload costs across Gemini 4 Argon, Gemini 3.1 Pro, and competitors like GPT-6 or Claude Opus?

👉 Launch the Interactive AI Cost Calculator
👉 View Gemini 4 Argon Specifications & Benchmarks