By TechZila AI Research Desk | Last checked: August 2026 | Reading time: ~10 minutes
Quick Take
- Release: Gemini 3.7 Flash launched August 13, 2026 — just 23 days after 3.6 Flash.
- Pricing: $0.75/1M input tokens, $3.75/1M output tokens through Dec 31, 2026 (50% off), then doubles to $1.50/$7.50 from Jan 1, 2027.
- Speed: 340 tokens/second output — fastest among 186 tested models, nearly 3x faster than GPT-5.6 Terra.
- Context: 1 million token input window, 65,536 token output limit.
- Key Gains: +9.2 points on FrontierCode (43.6% vs 34.4%), +16.3 points on DeepSWE (65.3% vs 49.0%).
- Best For: Coding, agentic workflows, long-horizon tasks, multimodal processing.
Google did something unusual in the AI race: it shipped a meaningfully better model without waiting for a new training run. Gemini 3.7 Flash went live on August 13, 2026 — only 23 days after Gemini 3.6 Flash — and the pitch is straightforward: better almost everywhere, at half the launch price.
August 2026: What's New
Gemini 3.7 Flash is Google's most intelligent Flash model to date, built for complex coding, agentic workflows, and reliable multi-step execution. It supports tunable thinking levels (low, medium, high), accepts text/image/video/audio/PDF inputs, and delivers text-only outputs with a 1M token context window.
Gemini 3.7 Flash: Key Numbers
What's Inside
What Is Gemini 3.7 Flash?
Gemini 3.7 Flash is Google's latest Flash-tier model, optimized for coding, agentic workflows, and multimodal processing. It's the third Flash model in 2026, following 3.5 Flash-Lite (July) and 3.6 Flash (July 21). Unlike typical model releases that trade cost for quality, 3.7 Flash improves on both fronts — at least temporarily.
The model accepts text, images, video, audio, and PDF inputs, with a massive 1,048,576-token input window and 65,536-token output ceiling. It supports configurable thinking levels (low, medium, high), letting developers tune latency versus intelligence based on task requirements.
| Specification | Value |
|---|---|
| Release Date | August 13, 2026 |
| Context Window | 1,048,576 input tokens |
| Output Limit | 65,536 tokens |
| Input Modalities | Text, Image, Video, Audio, PDF |
| Output Modality | Text only |
| Thinking Levels | Low, Medium (default), High |
| Speed | ~340 output tokens/second |
| Intelligence Index | 56 (Artificial Analysis) |
Pricing: 50% Discount (But There's a Catch)
Here's where things get interesting. Gemini 3.7 Flash launched at $0.75 per million input tokens and $3.75 per million output tokens — exactly half of Gemini 3.6 Flash's original $1.50/$7.50 pricing.
But here's the twist: on the same day 3.7 Flash launched, Google cut 3.6 Flash to the same $0.75/$3.75 introductory rate. So both models cost identical through 2026 — making the "50% off" claim technically accurate but context-dependent.
| Model | Through Dec 31, 2026 | From Jan 1, 2027 |
|---|---|---|
| Gemini 3.7 Flash | $0.75 input / $3.75 output | $1.50 input / $7.50 output |
| Gemini 3.6 Flash | $0.75 input / $3.75 output | $1.50 input / $7.50 output |
| Gemini 3.6 Flash (original) | N/A | $1.50 input / $7.50 output |
Additional pricing tiers include cached input at $0.075/1M tokens, cache storage at $0.50/1M tokens/hour, and batch inference at $0.375/$1.875 (input/output) through 2026.
Speed: 340 Tokens/Second
Speed is where Gemini 3.7 Flash genuinely stands out. Independent benchmark firm Artificial Analysis measured the model at 340.1 output tokens per second — ranking first among 186 tested models for generation speed.
For context, that's nearly 3x the output speed of GPT-5.6 Terra and GLM-5.2, which hover around 100-120 tokens/second. This makes 3.7 Flash ideal for latency-critical applications like real-time chat, incident response pipelines, and interactive coding assistants.
But there's a tradeoff: 3.7 Flash uses about 40% more output tokens than 3.6 Flash to complete the same tasks, averaging around 37,000 tokens per task. This puts it in the same range as Claude Fable 5 and Alibaba's Qwen3.8 Max, but it means actual cost-per-task may be higher than raw token pricing suggests.
Benchmarks: Coding & Agent Gains
Google's own benchmarks show significant improvements over 3.6 Flash, particularly in coding and agentic workflows:
| Benchmark | 3.7 Flash | 3.6 Flash | Improvement |
|---|---|---|---|
| FrontierCode 1.1 Main (Production code quality) | 43.6% | 34.4% | +9.2 points |
| DeepSWE v1.1 (Long-horizon software engineering) | 65.3% | 49.0% | +16.3 points |
| Code Arena — Web Development (Elo rating) | 1588 | 1538 | +50 Elo |
| Terminal-Bench 2.1 (Terminal tasks) | 85.8% | 78.0% | +7.8 points |
| AutomationBench (Business workflows) | 30.4% | 17.0% | +13.4 points |
| GDP.pdf (Document processing) | 34.0% | 22.0% | +12.0 points |
| OSWorld-2.0 (OS automation) | 47.9% | 33.8% | +14.1 points |
| 128K Long Context | 97.0% | 91.8% | +5.2 points |
The biggest gains are in long-horizon tasks (DeepSWE: +16.3 points) and business workflow automation (AutomationBench: +13.4 points). This suggests 3.7 Flash is particularly well-suited for multi-step coding projects and agentic workflows that require sustained reasoning.
Key Features & Capabilities
Gemini 3.7 Flash supports the full suite of Gemini API features, including:
Supported Capabilities
- Thinking modes: Low, Medium, High (minimal not supported)
- Caching: Supported for cost optimization
- Code execution: Supported
- Computer use: Supported (Preview)
- File search: Supported
- Function calling: Supported
- Grounding with Google Maps: Supported
- Search grounding: Supported
- Structured outputs: Supported
- URL context: Supported
- Batch API: Supported
- Flex inference: Supported
- Priority inference: Supported
Not Supported
- Audio generation: Not supported
- Image generation: Not supported
- Live API: Not supported
Input modalities include text, images (up to 3,000 per prompt), video (up to 45 minutes with audio, 1 hour without), audio (up to 8.4 hours or 1M tokens), and PDFs (up to 3,000 pages per file).
vs GPT-5.6 Terra & Claude Sonnet 5
On Artificial Analysis's composite Intelligence Index, all three models land within two points:
| Model | Intelligence Index | Speed (tok/s) | Price (through 2026) |
|---|---|---|---|
| GPT-5.6 Terra | 57 | ~110 | ~$2.50/$10.00 |
| Gemini 3.7 Flash | 56 | 340 | $0.75/$3.75 |
| Claude Sonnet 5 | 55 | ~100 | ~$3.00/$15.00 |
Gemini 3.7 Flash trades a single point of intelligence for 3x the speed and roughly 1/3 the blended cost of competitors. For coding and agentic workflows where speed and cost matter more than marginal intelligence gains, this is a compelling tradeoff.
Should You Switch from 3.6 Flash?
Quick Decision Guide
Switch to 3.7 Flash if: You need better coding performance (especially long-horizon tasks), faster response times, improved document processing, or agentic workflow capabilities. The identical pricing through 2026 makes this a no-brainer for new projects.
Stick with 3.6 Flash if: You're already in production with stable results, your tasks don't benefit from the coding/agent improvements, or you're sensitive to the 40% higher token usage per task. Both models cost the same through December, so there's no rush to migrate.
TechZila Analysis
The Core Truth
Gemini 3.7 Flash is Google's clearest statement yet: the Flash tier is for coding and agents, not just cheap inference. The 9-16 point benchmark gains in coding and automation aren't incremental — they're meaningful enough to change which tasks you'd trust the model with.
The Hidden Context
Releasing 3.7 Flash just 23 days after 3.6 Flash — and cutting 3.6 to the same price — suggests Google is racing to establish Flash as the default for coding before competitors catch up. The temporary 50% discount is a classic land-grab tactic: get developers hooked now, raise prices later.
TechZila Opinion
Frequently Asked Questions
When was Gemini 3.7 Flash released?
Gemini 3.7 Flash launched on August 13, 2026 — just 23 days after Gemini 3.6 Flash.
How much does Gemini 3.7 Flash cost?
Through December 31, 2026: $0.75 per million input tokens and $3.75 per million output tokens. From January 1, 2027: $1.50 input and $7.50 output (pricing doubles).
What is the context window for Gemini 3.7 Flash?
1,048,576 input tokens (1M context) with a 65,536 token output limit.
How fast is Gemini 3.7 Flash?
Approximately 340 output tokens per second — fastest among 186 models tested by Artificial Analysis, nearly 3x faster than GPT-5.6 Terra.
What thinking levels does 3.7 Flash support?
Low, Medium (default), and High. The "minimal" thinking level is not supported and returns an error.
Is Gemini 3.7 Flash better than 3.6 Flash?
Yes, particularly for coding (+9.2 points on FrontierCode), long-horizon software engineering (+16.3 points on DeepSWE), and business workflows (+13.4 points on AutomationBench). Both models cost the same through 2026.
What input types does 3.7 Flash accept?
Text, images (up to 3,000), video (up to 45 min with audio), audio (up to 8.4 hours), and PDFs (up to 3,000 pages). Output is text-only.
Last updated: August 16, 2026
0 Comments