GitHub Copilot: Gemini 3.6 Flash Model
GitHub CopilotGitHub Copilot now supports Gemini 3.6 Flash, Google's latest Flash-tier model, purpose-built for web and app development, coding, and longer-horizon agentic workflows. The model adds configurable reasoning effort and parallel tool use across complex tasks, and in early testing showed higher task-completion rates and up to 17% better token efficiency than Gemini 3.5 Flash. It rolls out to Copilot Pro, Pro+, Max, Business, and Enterprise plans across VS Code, Visual Studio, JetBrains, Xcode, Eclipse, the Copilot CLI, and the cloud agent, billed under usage-based pricing.
Key Takeaways
- Configurable reasoning effort lets Gemini 3.6 Flash scale its thinking depth to match task complexity, rather than applying uniform reasoning across simple and complex requests.
- Parallel tool use allows the model to coordinate multiple tool calls simultaneously, a meaningful upgrade for multi-step agentic coding sessions.
- Independent benchmarking cited by Google shows up to 65% fewer output tokens on DeepSWE-style coding tasks compared to Gemini 3.5 Flash.
- Output token pricing dropped from $9 to $7.50 per million tokens, making the model notably cheaper to run at scale.
- Enterprise and Business administrators must explicitly opt in via the "Gemini 3.6 Flash Preview" policy before their organizations can access the model, unlike individual-tier rollout.
- Hacker News commenters were split: some welcomed the price/efficiency gains, while others voiced skepticism about Google's ability to reliably provision capacity for the new tier.
Sources & Mentions
3 external resources covering this update
A New Flash-Tier Model Arrives in Copilot
GitHub has rolled out Gemini 3.6 Flash, Google's newest Flash-tier model, as a selectable option inside GitHub Copilot's model picker. The model is positioned specifically for web and app development, coding tasks, and extended agentic workflows — the kind of multi-step, tool-heavy sessions that have become increasingly common as Copilot's agent mode matures.
What's New Under the Hood
Gemini 3.6 Flash introduces configurable reasoning effort, letting the model dial its depth of reasoning up or down depending on task complexity, and adds support for parallel tool use so it can coordinate multiple tool calls at once during agentic workflows. According to GitHub's early testing, the model achieved higher task-completion rates than its predecessor, Gemini 3.5 Flash, while also being meaningfully more token-efficient — external benchmarking cited by Google put the reduction in output tokens at roughly 17% on the Artificial Analysis Index, with some coding-specific benchmarks like DeepSWE showing efficiency gains as high as 65%. That efficiency translated into a price cut as well: output pricing dropped from $9 to $7.50 per million tokens, while input pricing held steady at $1.50 per million tokens.
Where and How to Access It
Gemini 3.6 Flash is available to Copilot Pro, Pro+, Max, Business, and Enterprise subscribers, and can be selected from the model picker in Visual Studio Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, the Copilot CLI, and the GitHub Copilot cloud agent. Individual-tier users can pick the model immediately, but Business and Enterprise administrators need to explicitly enable the "Gemini 3.6 Flash Preview" policy in Copilot settings before members of their organization can use it. Because the rollout is gradual, availability may vary slightly by account in the initial days after launch. Usage is billed under standard usage-based billing at provider list pricing rates, consistent with GitHub's broader move to token-based billing across all Copilot plans.
Why It Matters
Flash-tier models are the workhorses for high-volume, latency-sensitive coding tasks, and this release continues a rapid cadence of GitHub adding third-party frontier models — alongside OpenAI and Anthropic options — directly into Copilot's model picker. For teams running frequent agentic sessions, the token-efficiency gains are the most tangible benefit, translating into lower per-session costs without sacrificing task-completion quality.