WayToClawEarn
High impactGoogle AI Blog

Google releases Gemini 3.5 Flash: pricing triples, reasoning capabilities significantly improved

Google officially released the Gemini 3.5 Flash model, whose performance is close to the previous generation Pro level, but the input/output prices have increased to $1.50/$9.00 per million tokens respectively. Developers need to re-evaluate usage scenarios. This article analyzes pricing changes, performance improvements, and adaptation suggestions.

WayToClawEarn EditorialPublished May 20, 2026Updated Aug 8, 2026

Editorial review of public sources · AI-assisted drafting. How we work · Original source

Core conclusion

On May 19, 2026, Google officially released the Gemini 3.5 Flash model. This is a major iteration of the Gemini series Flash product line. Its inference capabilities are close to the previous generation Pro level, but the price has also increased by about 3 times. For developers using Gemini APIs in automated workflows, this means re-evaluating cost structures and usage scenarios.

Key Points

  • Release time: 2026-05-19 -Affected objects: Automated pipelines and AI Agent workflows using Gemini API
  • Core changes: Reasoning ability greatly improved + price increased to $1.50/$9.00 (input/output)
  • Highlights: The output speed is extremely fast, it only takes ~1 second to generate 2000 tokens

Background and release information

Google directly skipped the official version of 3.0 and released 3.5 Flash after Gemini 2.5 Flash (end of 2025) and Gemini 3.0 Flash Preview (early 2026). This naming jump hints at a major refactoring of the code base layer - from community analysis, 3.5 Flash is most likely based on the MoE (Mixed Expert) architecture, running on TPU 8i, and achieving efficient services through RadixAttention-style batch inference.

Data source: Google official blog link and community discussion obtained by HN FireBase API (642 points, 400+ comments).

Key Impact

DimensionsChangeWhat it means to usRecommended actions
Inference qualityClose to the previous generation Pro levelFlash can do more thingsEvaluate whether Pro calls in existing automation processes can be downgraded to Flash
Input price$0.30→$1.50/ million (5x)Increased cost of processing large batches of dataPre-screen data and reduce useless token consumption
Output price$2.50→$9.00/ million (3.6x)Double the cost of long text generation scenariosControl the output length and switch thinking modes to reduce costs
Reasoning speed2000 tokens/clocks ≈1sExcellent interactive scene experienceSuitable for Chat UI, real-time Agent response
Tool callingCommunity testing shows that tool capabilities are decliningAdditional verification is required in the Agent frameworkAdd a fallback mechanism, or downgrade to 2.5 Flash for tool calling

Price comparison with other models

ModelInput ($/ million tokens)Output ($/ million tokens)Positioning
Gemini 2.5 Flash$0.30$2.50The previous generation’s best value for money
Gemini 3.0 Flash Preview$0.50$3.00Transitional version
Gemini 3.5 Flash$1.50$9.00The Year of Play
Gemini 2.5 Pro$1.25$10.00Pro-level reference system
DeepSeek V4~$0.25~$1.00Price Anchor

— Gemini pricing comparison chart

Adaptation suggestions

Based on the pricing and quality changes of Gemini 3.5 Flash, the following are specific recommendations for different scenarios:

Scenario 1: AI automated content production

  • For quality-sensitive tasks (such as long text generation), give priority to 3.5 Flash, the quality is close to Pro and the price is lower
  • For cost-sensitive tasks (such as batch summarization), downgrade to 2.5 Flash, which is of sufficient quality and 5x cheaper
  • For tool calling tasks, first run an A/B test to confirm whether the tool use of 3.5 Flash meets the standards.

Scenario 2: AI Agent/Coding Assistance

  • Claude Code + DeepSeek V4 combination is still the most cost-effective encoding solution (10x cheaper)
  • Gemini 3.5 Flash is suitable for scenarios that require extremely fast response speed (interactive Chat UI, real-time Agent)
  • It is recommended to add exponential backoff and observable logs to failed retries to track the success rates and costs of different models

Scenario 3: Content distribution and SEO optimization

  • High-throughput content rewriting task: If you used 2.5 Pro before, you can downgrade to 3.5 Flash for testing
  • If it is a new project, it is recommended to start with 2.5 Flash, and then upgrade to 3.5 Flash if the quality is not enough

Tip: 3.5 Flash consumes quota very quickly in the Google AI Pro plan (some users reported that the full-day quota can be used after two rounds of conversations). Be sure to test the quota consumption speed before official use.

Extended thinking: Flash model pricing trends

From 2.5 Flash to 3.5 Flash, Google's Flash product line has achieved a "capability jump + price reshaping." In the past, the core selling point of Flash was "cheap and useful", but now the selling point of 3.5 Flash has become "close to Pro quality + Flash speed".

The impact of this on content automation operations:

  • It used to be a choice between Pro (expensive but powerful) and Flash (cheap but weak)
  • Now 3.5 Flash provides a "strong enough" choice, but the price has also reached the middle range
  • Future AI pipelines require more sophisticated model routing: cheap models (DeepSeek V4/2.5 Flash) for simple tasks, 3.5 Flash for complex tasks, and Pro for the most critical tasks

— AI model decision flowchart

Tool entry

AI tools and models covered in the text: Gemini, Google AI, DeepSeek, Claude, Claude Code, OpenAI, n8n

Reference sources

Related reading

Want to learn how? See our full tutorial: How to build an automated coding assistant with Gemini 3.5 Flash API: a complete 30-minute tutorial

Some people have successfully implemented it: He Built an AI Automation Stack with Claude + n8n — $4K to $12K/mo in 6 Months

View source →

Disclaimer: this site shares educational insights only, for inspiration and reference. No outcome guarantee; external execution and decisions are your own responsibility.