Google’s Gemini 3.6 Flash is positioned around a combination that matters to developers: lower token consumption and stronger results on complex work. The company says the model uses 17% fewer tokens than Gemini 3.5 Flash across a range of tasks.
The reduction comes as Google expands the model’s role in multistep work involving coding, knowledge tasks, and computer-interface operation. Gemini 3.6 Flash is also said to require fewer reasoning steps and fewer additional tools when handling such workloads.
Better Results on Code Tasks
The coding gains are among the most notable changes in the new model. According to Liputan6.com, citing 9to5Google, Gemini 3.6 Flash is more accurate at coding work and requires fewer code edits and repeated repair cycles.
Its DeepSWE code-quality score rose from 37% to 49%. Performance on machine learning research also increased from 49.7% to 63.9%.
| Benchmark | Earlier Score | Gemini 3.6 Flash |
|---|---|---|
| DeepSWE | 37% | 49% |
| Machine learning research | 49.7% | 63.9% |
| GDPval-AA | 1349 | 1421 |
| OSWorld-Verified | 78.4% | 83% |
Gemini 3.6 Flash achieved a score of 1421 on GDPval-AA, compared with 1349 for the comparison model. Its knowledge cutoff was also updated from January 2025 to March 2026.
On OSWorld-Verified, which measures computer-interface operation, the score increased from 78.4% to 83%. The result supports Google’s focus on tasks that go beyond text generation and involve action across computer environments.
Pricing for Multistep Workloads
Google has set the price of Gemini 3.6 Flash at USD 1.50 per 1 million input tokens and USD 7.50 per 1 million output tokens. The model was introduced after an earlier release at I/O 2026, incorporating feedback from developers and customers.
| Model | Primary Focus | Input Price | Output Price |
|---|---|---|---|
| Gemini 3.6 Flash | Multistep tasks, coding, computer interfaces | USD 1.50 per 1 million tokens | USD 7.50 per 1 million tokens |
| Gemini 3.5 Flash-Lite | Agent-based search, document processing | USD 0.30 per 1 million tokens | USD 2.50 per 1 million tokens |
Google is also offering Gemini 3.5 Flash-Lite for workloads where speed and low latency take priority. It is intended for agent-based search and document-processing tasks.
Flash-Lite’s Terminal-Bench 2.1 score for programming and agent-based tasks climbed from 31% to 54%. Its long-context score on GDM-MRCR v2 rose from 60.1% to 72.2%.
The lighter model improved from 65.1% to 74.0% on OSWorld-Verified. It also reached 54.2% on SWE-Bench Pro, above the 49.6% recorded by Gemini 3 Flash.
Cyber Model and Availability
Google has prepared Gemini 3.5 Flash Cyber to identify and repair security vulnerabilities in code. Its Flash model performance and efficiency are intended to support large-scale detection, validation, and patching of code flaws.
Google’s CodeMender tool uses multiple Gemini 3.5 Flash Cyber agents. Early access to that model has been provided to governments and trusted partners through a small-scale trial program.
Gemini 3.6 Flash and Gemini 3.5 Flash-Lite are publicly available through the Gemini app. Developers can also access them through Google Antigravity, AI Studio, and Android Studio, while Gemini 3.5 Pro remains in testing with partners.