On July 21, 2026, Google introduced three new additions to its Gemini model family — Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber. Gemini 3.6 Flash is the new flagship "workhorse" Flash model, offering stronger coding, reasoning, and multimodal performance while using 17% fewer output tokens than Gemini 3.5 Flash. Gemini 3.5 Flash-Lite is the fastest and most affordable model in the lineup, built for high-throughput agentic workloads. Gemini 3.5 Flash Cyber is a specialized security model deployed inside Google's CodeMender agent to detect and patch software vulnerabilities. All three models are designed around one shared goal: making it practical to run AI agents in production at scale.
This post breaks down what each model does differently, how they perform against their predecessors on independent benchmarks, what they cost, and where you can start using them today.
This article covers:
What Gemini 3.6 Flash is and why it matters
Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber explained.
Performance benchmarks, token efficiency, and real-world improvements
Gemini 3.6 Flash vs. Gemini 3.5 Flash vs. Gemini 3.5 Flash-Lite
Pricing, availability, release date, and how to access the models
Built-in computer use, agentic AI capabilities, and multimodal features
Gemini 3.5 Flash Cyber, CodeMender, and Google's AI security strategy
Safety improvements, enterprise adoption, and production use cases
Gemini 3.5 Pro, Gemini 4 roadmap, and what's coming next
FAQs, limitations, and whether Gemini 3.6 Flash is worth using
What Is Gemini 3.6 Flash?
Gemini 3.6 Flash is Google's latest mid-tier Gemini model, positioned as the direct successor to Gemini 3.5 Flash. It is built specifically for developers and enterprises running agentic systems — AI workflows that involve multiple reasoning steps, tool calls, and long chains of automated decision-making. Rather than simply being "smarter" in the abstract, Gemini 3.6 Flash is engineered around three practical priorities that matter when AI moves from a demo into a real product: token efficiency, lower latency, and consistent reliability.
According to Google, Gemini 3.6 Flash consumes roughly 17% fewer output tokens than Gemini 3.5 Flash on the Artificial Analysis Index, and on certain coding-heavy benchmarks such as DeepSWE (built by Datacurve), that efficiency gain climbs as high as 65%. In practice, this means the model reaches the same or better quality outcomes while generating substantially less "wasted" text, which directly reduces the cost of running long, multi-step agent tasks.
Meet the Full Lineup: 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
Google's announcement bundles three distinct models, each aimed at a different part of the agentic-AI stack:
Gemini 3.6 Flash — the general-purpose workhorse model for coding, knowledge work, and multimodal tasks like document parsing and chart analysis.
Gemini 3.5 Flash-Lite — the fastest, most cost-efficient model in the 3.5 generation, purpose-built for high-volume, latency-sensitive workloads such as agentic search and bulk document processing.
Gemini 3.5 Flash Cyber (in CodeMender) — a narrow, specialized model fine-tuned exclusively for finding and fixing software vulnerabilities, deployed as part of Google's CodeMender security agent rather than sold as a general-purpose API model.
Together, these releases reflect a broader industry shift: instead of chasing a single "biggest" model, Google is building a tiered family where each model is optimized for a specific cost, speed, and capability trade-off — a pattern also playing out at OpenAI, Anthropic, and Meta as AI shifts from chat assistants toward autonomous, task-executing agents.
Performance Benchmarks: How Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber Measure Up
Independent and internal benchmark results give a clearer picture of what "better" actually means for each model. Here is a consolidated view of the published figures.
Gemini 3.6 Flash benchmark results (vs. Gemini 3.5 Flash)
Benchmark | What It Measures | 3.5 Flash | 3.6 Flash |
DeepSWE (Datacurve) | Real-world software engineering precision | 37% | 49% |
MLE Bench | Machine-learning research tasks | 49.7% | 63.9% |
OSWorld-Verified | Computer-use / operating-system task execution | 78.4% | 83.0% |
GDPval-AA v2 | Real-world knowledge-work quality | 1,349 | 1,421 |
Output token usage | Efficiency (lower is better) | Baseline | 17% fewer tokens |
Gemini 3.5 Flash-Lite benchmark results (vs. Gemini 3.1 Flash-Lite)
Benchmark | What It Measures | 3.1 Flash-Lite | 3.5 Flash-Lite |
Terminal-Bench 2.1 | Coding and agentic terminal tasks | 31% | 54% |
GDM-MRCR v2 | Long-context comprehension | 60.1% | 72.2% |
GDPval-AA v2 | Real-world task execution quality | 642 | 1,140 |
Output speed (Artificial Analysis) | Tokens generated per second | Slower | 350 tokens/sec |
Notably, Gemini 3.5 Flash-Lite doesn't just outperform its direct predecessor — Google reports it also beats the larger Gemini 3 Flash on several agentic and coding evaluations, including SWE-Bench Pro (54.2% vs. 49.6%) and OSWorld-Verified (74.0% vs. 65.1%). That makes it a viable, faster replacement for teams currently running workloads on either 2.5 Flash or 3 Flash.
Gemini 3.5 Flash Cyber benchmark performance
Inside CodeMender, multiple Gemini 3.5 Flash Cyber agents collaborate to produce a single combined vulnerability report. On the widely used CyberGym benchmark, Google states that this multi-agent Flash Cyber configuration reaches performance competitive with frontier-class models — a meaningful result given that Flash Cyber is a smaller, cheaper, more efficient model than the large frontier systems it's being measured against.
Gemini 3.6 Flash vs. Gemini 3.5 Flash: What's Actually Different
The jump from Gemini 3.5 Flash to Gemini 3.6 Flash isn't a marginal update — it touches quality, efficiency, and price at the same time, which is unusual, since these three factors typically trade off against one another.
Efficiency: Gemini 3.6 Flash needs fewer reasoning steps and fewer tool calls to complete the same multi-step task, and it produces about 17% fewer output tokens on average, with some coding benchmarks showing reductions of up to 65%.
Quality: Across coding (DeepSWE), machine-learning research (MLE Bench), computer use (OSWorld-Verified), and general knowledge work (GDPval-AA v2), Gemini 3.6 Flash posts clear gains over Gemini 3.5 Flash. Google specifically calls out improvements in document parsing, chart and data analysis, and report drafting — capabilities that enterprise customers such as Hebbia and Harvey have highlighted as particularly strong in real deployments.
Price: Despite the quality improvements, Gemini 3.6 Flash is priced lower than its predecessor at $1.50 per 1 million input tokens and $7.50 per 1 million output tokens. Combined with the token-efficiency gains, Google says this meaningfully lowers the total cost of running an agentic task from start to finish — not just the per-token rate.
New built-in capability: Computer use — the ability for the model to directly operate software interfaces — is now available as a built-in, client-side tool through both the Gemini API and Gemini Enterprise, rather than requiring separate custom tooling.
In short: Gemini 3.6 Flash is positioned as a straightforward upgrade path for any team currently building on Gemini 3.5 Flash, offering better results at a lower cost with no architectural changes required on the developer's end.
Gemini 3.5 Flash-Lite vs. Gemini 3.1 Flash-Lite: How Big Is the Generational Leap?
If Gemini 3.6 Flash is about balancing power and efficiency, Gemini 3.5 Flash-Lite is about pure speed and cost at scale — and the gap between it and its predecessor, Gemini 3.1 Flash-Lite, is substantial.
Speed: Gemini 3.5 Flash-Lite is the fastest model in the 3.5 generation, generating 350 output tokens per second, according to Artificial Analysis. That directly addresses latency-sensitive use cases like live agentic search, chat-based customer support, and real-time document triage.
Coding and agentic tasks: On Terminal-Bench 2.1, Gemini 3.5 Flash-Lite nearly doubles the score of Gemini 3.1 Flash-Lite (54% vs. 31%).
Long-context understanding: On GDM-MRCR v2, a benchmark for long-context comprehension, Gemini 3.5 Flash-Lite scores 72.2% compared to 60.1% for Gemini 3.1 Flash-Lite — an important gain for tasks like analyzing lengthy contracts, transcripts, or multi-document research sets.
Real-world task quality: On GDPval-AA v2, which evaluates real-world task execution, Gemini 3.5 Flash-Lite nearly doubles its predecessor's score (1,140 vs. 642).
Pricing: Gemini 3.5 Flash-Lite is priced at $0.30 per 1 million input tokens and $2.50 per 1 million output tokens, making it Google's most budget-friendly production-grade model in the current lineup.
Developers can also tune Gemini 3.5 Flash-Lite's "thinking level" depending on the workload — using minimal or low thinking settings for high-volume, low-latency tasks, or dialling up reasoning depth for more complex, multi-step subagent workflows. Like Gemini 3.6 Flash, it also ships with computer use as a built-in tool.
Real-world adoption: Early users of Gemini 3.5 Flash-Lite — including Ashler, Palo Alto Networks, and Ramp — have pointed to its combination of speed, intelligence, and cost efficiency as the reason it fits into large-scale agentic and data-processing pipelines where every millisecond and every token counts.
Gemini 3.5 Flash Cyber in CodeMender: AI-Powered Vulnerability Detection
Cybersecurity is one of the clearest examples of where AI capability has started to outpace traditional defence: automated systems can now discover software vulnerabilities faster than human teams — or even most existing tools — can patch them. Google's answer to this imbalance is Gemini 3.5 Flash Cyber, a specialized model built on top of Gemini 3.5 Flash and fine-tuned specifically for identifying, validating, and fixing security vulnerabilities in code.
Rather than being released as a standalone general-purpose API model, Gemini 3.5 Flash Cyber is deployed inside CodeMender, Google's dedicated code-security agent. Within CodeMender, multiple Flash Cyber agents work in parallel and combine their findings into a single, unified vulnerability report — an orchestration approach that lets a smaller, cheaper model achieve results competitive with far larger frontier systems on the CyberGym benchmark, at a fraction of the cost per token.
Because vulnerability-detection technology is inherently dual-use — the same capability that helps defenders patch a flaw could, in the wrong hands, help an attacker exploit one — Google is taking a deliberately narrow rollout approach. Gemini 3.5 Flash Cyber in CodeMender will be made available exclusively to governments and trusted partners, through a limited-access pilot program, rather than opened broadly through the public API. The stated goal is to give defensive security teams a head start on discovering and closing critical vulnerabilities before attackers can exploit them, while limiting the risk of the same technology being misused offensively.
Built-In Safety: How Google Is Hardening Gemini 3.6 Flash
Alongside the performance and efficiency improvements, Gemini 3.6 Flash ships with strengthened protections under Google's Frontier Safety Framework, specifically targeting two high-risk misuse categories: Chemical, Biological, Radiological, and Nuclear (CBRN) risks and cyber-offence misuse. These safeguards are designed to make the model substantially harder to jailbreak or manipulate into producing harmful outputs in those domains.
At the same time, Google says it deliberately trained the model to avoid over-refusing legitimate, beneficial requests — a balancing act that reflects a common tension in frontier AI development: locking a model down too aggressively can make it frustrating or useless for the very developers and researchers it's meant to serve, while locking it down too little creates real-world risk. Full technical detail on these protections is documented in the official Gemini 3.6 Flash model card published by Google DeepMind.
Rollout Timeline: Release Date, Availability, and Pricing
Release date: Google announced Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber on July 21, 2026, in a post authored by Tulsee Doshi, Senior Director of Product Management for the Gemini team.
Where to access Gemini 3.6 Flash and Gemini 3.5 Flash-Lite:
Developers: Available now through the Gemini API via Google AI Studio and Android Studio. Gemini 3.6 Flash is additionally available in Google Antigravity, Google's agentic development environment.
Enterprises: Available through the Gemini Enterprise Agent Platform, with Gemini 3.6 Flash also accessible directly inside the Gemini Enterprise app.
Everyday users: Both models are rolling out through the consumer Gemini app, and Gemini 3.5 Flash-Lite is additionally rolling out inside Google Search.
Pricing at a glance:
Model | Input Price (per 1M tokens) | Output Price (per 1M tokens) |
Gemini 3.6 Flash | $1.50 | $7.50 |
Gemini 3.5 Flash-Lite | $0.30 | $2.50 |
Gemini 3.5 Flash Cyber | Not publicly priced (limited-access pilot via CodeMender only) | Not publicly priced |
For teams evaluating which model fits a given workload, the general guidance emerging from Google's own framing is straightforward: reach for Gemini 3.6 Flash when task quality and reasoning depth matter most, choose Gemini 3.5 Flash-Lite when speed and cost-per-request are the priority at high volume, and expect Gemini 3.5 Flash Cyber to remain a specialized, access-controlled tool for vulnerability research rather than a general development model.
What Comes Next: Gemini 3.5 Pro and Gemini 4
Google confirmed that Gemini 3.5 Pro is currently being tested with select partners, with plans to make it broadly available once testing is complete — though no firm release date has been given. Looking further ahead, Google stated that its research team has already begun its most ambitious pre-training run to date, aimed at building Gemini 4, signalling that the Flash-tier releases covered here are very much an interim step in a much longer product roadmap rather than an endpoint.
The Bottom Line
Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber represent a coordinated push from Google to make production AI agents cheaper to run, faster to respond, and more dependable at scale — without asking developers to compromise on output quality. Gemini 3.6 Flash raises the ceiling on coding and knowledge-work performance while cutting token costs. Gemini 3.5 Flash-Lite pushes the floor on speed and affordability for high-volume workloads. And Gemini 3.5 Flash Cyber, tightly scoped inside CodeMender, shows how a smaller specialised model can compete with frontier systems on a narrow, high-stakes task like vulnerability detection.
For teams building agentic products today, the practical takeaway is simple: Gemini 3.6 Flash is a low-friction upgrade from Gemini 3.5 Flash, Gemini 3.5 Flash-Lite is worth testing against current 2.5 Flash or 3 Flash deployments for high-throughput workloads, and Gemini 3.5 Flash Cyber signals where Google's security research is headed next, even if it isn't broadly available yet.
FAQs
What is Gemini 3.6 Flash used for?
Gemini 3.6 Flash is designed for building AI agents that need strong coding ability, multimodal understanding (documents, charts, images), and reliable multi-step reasoning, while keeping token usage and cost lower than its predecessor, Gemini 3.5 Flash.
Is Gemini 3.6 Flash better than Gemini 3.5 Flash?
Yes. Gemini 3.6 Flash scores higher across coding (DeepSWE), machine-learning research (MLE Bench), computer-use tasks (OSWorld-Verified), and knowledge work (GDPval-AA v2), while using roughly 17% fewer output tokens and costing less per million tokens.
What is Gemini 3.5 Flash-Lite best for?
Gemini 3.5 Flash-Lite is built for high-throughput, latency-sensitive workloads such as agentic search, large-scale document processing, and real-time applications, where speed (350 tokens/sec) and low cost matter more than maximum reasoning depth.
How much does Gemini 3.6 Flash cost?
Gemini 3.6 Flash is priced at $1.50 per 1 million input tokens and $7.50 per 1 million output tokens.
How much does Gemini 3.5 Flash-Lite cost?
Gemini 3.5 Flash-Lite is priced at $0.30 per 1 million input tokens and $2.50 per 1 million output tokens.
What is Gemini 3.5 Flash Cyber in CodeMender?
Gemini 3.5 Flash Cyber is a specialised cybersecurity model built on Gemini 3.5 Flash and fine-tuned to find and fix code vulnerabilities. It runs inside CodeMender, Google's AI code-security agent, where multiple Flash Cyber agents collaborate to produce a unified vulnerability report. It is being rolled out exclusively to governments and trusted partners through a limited-access pilot rather than the public API.
Can I access Gemini 3.5 Flash Cyber through the public API?
No. Unlike Gemini 3.6 Flash and Gemini 3.5 Flash-Lite, Gemini 3.5 Flash Cyber is not a general-availability API model. It is being deployed selectively through CodeMender to governments and trusted security partners due to the dual-use risk of vulnerability-detection technology.
Does Gemini 3.5 Flash-Lite outperform Gemini 3 Flash?
On several benchmarks, yes. Google reports Gemini 3.5 Flash-Lite beat the larger Gemini 3 Flash on SWE-Bench Pro (54.2% vs. 49.6%) and OSWorld-Verified (74.0% vs. 65.1%), despite being a smaller, faster, cheaper model.
When is Gemini 3.5 Pro coming out?
As of the July 21, 2026 announcement, Gemini 3.5 Pro is in the partner testing phase, with broad availability planned once testing concludes. Google has not published a specific public release date.
Where can I try Gemini 3.6 Flash and Gemini 3.5 Flash-Lite today?
Developers can access both through Google AI Studio, Android Studio, and the Gemini API; enterprises through the Gemini Enterprise Agent Platform and app; and general users can access through the Gemini app, with Gemini 3.5 Flash-Lite also rolling out in Google Search.