gemini 3 6 flash Google has updated its Gemini lineup again, rolling out Gemini 3.6 Flash, introducing a cybersecurity-focused model, and offering a first look at its longer-term plans for Gemini 4. The company also confirmed that Gemini 3.5 Pro remains stuck in testing, despite earlier expectations that it would arrive in June.
gemini 3 6 flash
Gemini 3.6 Flash replaces 3.5 Flash
The most immediate change is Gemini 3.6 Flash, which Google is positioning as a faster, cheaper, and slightly more capable successor to Gemini 3.5 Flash. The company says the new model improves coding and multimodal performance, while also responding to user feedback on the previous release.
That feedback appears to have mattered. Google’s earlier 3.5 Flash model was presented as a highly efficient option, but it did not fully meet expectations for code generation. In practical terms, 3.6 Flash is meant to keep the same efficiency-first approach while narrowing the quality gap.
Google highlighted benchmark gains to support that pitch:
- DeepSWE coding test: 49 percent for Gemini 3.6 Flash, up from 37 percent for 3.5 Flash
- OSWorld computer-use test: 83 percent, compared with 78.4 percent for 3.5 Flash
- Token usage: about 17 percent fewer tokens than 3.5 Flash
The model also adds computer use as a standard feature in the Gemini API. Google says that should help with agentic workflows, where models complete multi-step tasks with fewer errors and fewer tokens. Lower token use matters because it directly affects operating costs for both developers and Google.
Pricing reflects that focus. Gemini 3.6 Flash costs $1.50 per 1 million input tokens and $7.50 per 1 million output tokens. That is the same input price as 3.5 Flash, but output tokens are cheaper than the previous $9 per 1 million.
Flash Lite gets faster, and Flash Cyber arrives
Google is not stopping at the main Flash tier. It also unveiled Gemini 3.5 Flash Lite and Gemini 3.5 Flash Cyber, extending the 3.5 branch even as 3.6 Flash becomes the new default in other places.
Flash Lite is now Google’s most efficient modern AI model, according to the company, with throughput of 350 tokens per second. Google says it is designed for scaling agentic systems without driving up costs. Benchmark results suggest it performs close to frontier models from roughly a year ago, while remaining relatively inexpensive.
Its pricing is set at $0.30 per 1 million input tokens and $2.50 per 1 million output tokens. That makes it more expensive than the earlier 3.1 Flash Lite on both counts, which Google priced at $0.25 and $1.50.
Google says Gemini 3.5 Flash Lite is available to developers and in the Gemini app. The company also says it will appear frequently in Google Search, where its speed could make it a good fit for AI Overviews.
The other new model, Gemini 3.5 Flash Cyber, is Google’s first LLM tuned specifically for cybersecurity. The company says it performs almost as well at finding and fixing security issues as the much larger and more expensive Claude Mythos, while keeping the efficiency profile of a Flash model.
Google is also explicit about the risks. Because cybersecurity models can be used both defensively and offensively, the company says Flash Cyber is too dangerous for broad public release. Instead, it will launch as a limited pilot inside Google DeepMind’s CodeMender agent, which is available only to trusted partners and governments.
What happened to Gemini 3.5 Pro?
One of the bigger unanswered questions is Gemini 3.5 Pro. Google previously said at I/O that the model would launch in June, but that never happened. Now the company says only that the model is in testing with unnamed partners and will ship “as soon as it’s ready.”
Google’s silence may reflect pressure around model quality. Earlier reporting suggested that 3.5 Pro was delayed because it could not match competitors in coding performance. In the latest update, Google did not offer a reason for the delay, nor did it commit to a launch date.
That matters because 3.5 Pro is supposed to be Google’s flagship model, positioned to compete with GPT 5.6 and Claude Fable/Sonnet 5. For now, though, the company is focusing attention on the Flash family and on more specialized products.
Gemini 4 is already in pre-training
Google’s roadmap also includes a brief mention of Gemini 4. The company says pre-training has already started, and it describes the process as more ambitious than its previous AI efforts. Beyond that, there are no specifics on timing, capabilities, or whether additional 3.x releases will appear before Gemini 4 arrives.
That leaves Google’s immediate AI strategy looking split across several tracks: incremental efficiency gains for general-purpose models, a guarded rollout for cybersecurity work, and a still-delayed flagship model that has yet to leave testing.
What changes now for users and developers
For most users, the practical shift is that Gemini 3.6 Flash will start rolling out in the API today and will replace 3.5 Flash in the Gemini app. Developers get a cheaper output rate and improved coding and computer-use performance. Users of Google Search may also see Flash Lite more often, which could improve response speed in AI Overviews.
In short, Google’s latest update is less about a single headline model than about a broader reshuffle: the company is pushing efficiency harder, adding specialized tools for security, and keeping its flagship plans under wraps while Gemini 4 begins pre-training.
Explore more: Blog Our Services Contact Us
Source: Original report
Was this helpful?
Last Modified: July 22, 2026 at 6:38 pm
0 views
