Google AI Pricing (August 2026)

WeatherNext pricing and access

WeatherNext is not a Gemini API model. Google has released WeatherNext 2, WeatherNext Cyclones, and a smaller Mini checkpoint as research code and model weights. Code and notebooks use Apache 2.0, while other repository materials use CC BY 4.0. There is no published Google per-token, per-run, or per-forecast model price, so WeatherNext does not appear as a zero-dollar row in our canonical pricing API.

The official Mini notebook recommends a free Colab v5e-1 runtime. Larger checkpoints require a v5p TPU or sufficient GPU memory; Google says non-Mini models need H100-class memory on GPU. Operational costs can also include initial-condition data, storage, transfer, orchestration, and forecast verification. Daily WeatherNext outputs are available through BigQuery, Earth Engine, and Cloud Storage, where normal platform usage charges may apply.

WeatherNext is also outside the current 49-task language-model Labs harness. A valid comparison needs pinned atmospheric inputs, forecast verification data, cyclone-specific skill metrics, ensemble-size controls, and accelerator-hour logs. See the WeatherNext pricing and cyclone-model analysis for the full cost boundary and explicit Labs blocker.

Managed Agents pricing

Managed Agents in the Gemini API now use Gemini 3.6 Flash by default. Google bills all input, output, intermediate input, and reasoning tokens at the selected model's standard rates, with normal tool fees on top. CPU, memory, and sandbox execution are free during the preview period. Free-tier projects can also try Managed Agents within Google's quotas.

The live pricing table on this page shows the current Gemini 3.6 Flash rates. Autonomous loops can generate many turns, so use the new token ceiling and estimate cost per completed workflow. See the Managed Agents pricing and hooks guide for rollout advice and the exact billing boundary.

Antigravity CLI also appears in a community omp setup as a separate search tool behind DeepSeek V4 Flash and GPT-5.6 Luna. Read the oh-my-pi cost and setup guide for the quota boundary and end-to-end Labs blocker.

Google's Gemini API stands out in July 2026 for a large production context window, competitive model pricing, and a managed-agent preview with no separate environment-compute fee. Gemini 3.6 Flash is now the default for the Antigravity agent, while Gemini 3.1 Pro remains the long-context flagship. The April 1 free-tier reshuffle removed free Pro-model access but kept Flash models available for prototyping.

Gemini 3.x (Current Production)

Google's latest generation is now GA and the recommended production tier:

• Gemini 3.6 Flash ($1.50 input / $7.50 output) — Current Flash leader and default for the Antigravity managed agent.
• Gemini 3.5 Flash-Lite ($0.15 input / $0.60 output) — Lowest-cost model supported by Managed Agents.
• Gemini 3.1 Pro ($2.00 / $12.00 per 1M under 200K; $4.00 / $18.00 above) — Current flagship. Paid-only as of April 1, 2026. 2M-token context window.
• Gemini 3 Pro ($2.00 input / $12.00 output) — Stable flagship alternative. Paid-only.
• Gemini 3 Flash ($0.50 input / $3.00 output) — Earlier Flash generation. Retains free tier with reduced quota.
• Gemini 3.1 Flash-Lite ($0.25 input / $1.50 output) — Cheapest Tier-1 budget model. Retains free tier with reduced quota.

Gemini 2.5 (Legacy — Paid Tier Only)

The 2.5 family remains available but is moving to legacy status. All 2.5 models are now paid-tier only.

• Gemini 2.5 Pro ($1.25 input / $10.00 output) — Legacy flagship. Still competitive on price but superseded by 3.x.
• Gemini 2.5 Flash ($0.30 input / $2.50 output) — Legacy mid-tier. Paid-only after April 1, 2026.

Free Tier: What Changed April 1, 2026

Google tightened the free tier significantly on April 1, 2026. The new state:

• Pro-tier models (3.1 Pro, 3 Pro, 2.5 Pro): free tier removed. Paid-only.
• Gemini 3 Flash: free tier retained with reduced daily quota.
• Gemini 3.1 Flash-Lite: free tier retained with reduced daily quota.

Google still has the most generous free tier overall — Flash access at scale is unmatched — but the "free Pro" era is over. For flagship-class access without payment, your only remaining option is Claude's $5 trial credits.

Full breakdown: What changed April 1 and three ways to keep prototyping cheaply →

Context Caching: 75-90% Savings

Gemini's context caching is exceptionally aggressive. Gemini 2.5 Pro cached input drops from $1.25 to $0.13/M — a 90% discount. For applications with repeated system prompts or reference documents, this can slash your bill dramatically.

How Gemini Compares

At the flagship tier, Gemini 3.1 Pro ($2.00/M input) matches GPT-5.4 on price and undercuts Claude Opus 4.8 ($5.00/M) by 60%. Gemini still wins on context window — 2M tokens vs Opus 4.8's 1M and GPT-5.4's 270K — and on vision resolution.

At the budget tier, Gemini 3.1 Flash-Lite ($0.25/M input) is the cheapest mainstream model from a Tier-1 provider in 2026. Only DeepSeek's cached pricing is lower, at the cost of non-Tier-1 operational maturity.

For detailed head-to-head benchmarks and cost scenarios, see our flagship comparison guide.