Google DeepMind has released three new Gemini Flash models that are all tuned for cheaper and faster AI agents. Meanwhile, the flagship Gemini 3.5 Pro that developers have waited on since Feb
Google DeepMind has released three new Gemini Flash models that are all tuned for cheaper and faster AI agents.
Meanwhile, the flagship Gemini 3.5 Pro that developers have waited on since February is still stuck in partner testing.
Google has released three new models, each with a specific job. All three new releases are part of the “Flash” family, which means they are lightweight and efficient compared to the biggest, most powerful AI models.
Gemini 3.6 Flash, the successor to Gemini 3.5 Flash, is the main “workhorse” model for general use. Google says it uses up to 17% fewer output tokens, making it more efficient. It is also better at coding and reasoning. The price is $1.50 for every million input tokens and $7.50 for every million output tokens.
Google said customers, including Hebbia and Harvey, found 3.6 Flash strong at document parsing, chart analysis, and report drafting.
Gemini 3.5 Flash-Lite is a model used for high-volume jobs where speed and cost are more important than maximum power. It is the cheapest model, costing $0.30 per million input tokens and $2.50 per million output tokens, and runs at 350 output tokens per second. Both 3.6 Flash and Flash-Lite are available now in Google’s developer tools and apps.
Gemini 3.5 Flash Cyber is a specialized cybersecurity model made to find and fix software vulnerabilities. It uses CodeMender, a tool Google showed off in 2025.
CodeMender calls the Flash Cyber model many times to scan a lot of code and then puts the findings into one report. In a test on the V8 JavaScript engine, Flash Cyber found 55 confirmed vulnerabilities. For comparison, Gemini 3.5 Flash found 47, and Anthropic’s Claude Opus 4.6 found 36.
Gemini 3.5 Flash Cyber won’t be released to the public right away. For now, due to safety concerns, it is only available to governments and trusted partners through a limited pilot program.
Google wants to give “frontline defenders a head start” in fixing bugs before they can be used for attacks. They are also trying to prevent a “broader misuse” of the technology. The model is a direct competitor to Anthropic’s more expensive cybersecurity model, Mythos.
Google promised back in May that 3.5 Pro would be released in June, but now in July, the product is yet to launch. Some reports say the model did not meet internal performance targets, especially for coding, while others say the delay is because of technical problems, and the product had to be rebuilt from scratch.
In the meantime, OpenAI has launched new models like GPT-5.5 and parts of GPT-5.6. Anthropic has released Claude Opus 4.8, Claude Sonnet 5, and Claude Fable 5.
Google says Gemini 3.5 Pro is currently being tested with partners and will be available “soon,” but no one knows when exactly that will be. The company has also started its “most ambitious pre-training run yet” for the next model, Gemini 4, but that is still a long way off.
Cryptopolitan previously covered Grok 4.5’s pricing pitch against Claude, explaining that a cost-per-task logic has changed how engineering teams pick their tools. Similarly, in this situation, rather than focus on making AI models smarter, companies are creating cheaper models that meet their needs.
Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.
