Gemini 3.8 Flash Pricing: How It Stacks Up for Coding

Google has taken Gemini 3.8 Flash generally available in the Gemini API, its third Flash-tier release in roughly six weeks and the direct successor to Gemini 3.7 Flash. If you are choosing a model for agentic coding work and searching for gemini 3.8 flash pricing before you commit a workflow to it, the short version is that Google kept the price where it was and pushed the performance up, at least on paper.

What Actually Shipped

Gemini 3.8 Flash (gemini-3.8-flash) is what Google calls its "most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows." It carries the same footprint as recent Flash releases: a 1,048,576 token input context window, a 65,536 token output ceiling, and multimodal input across text, image, video, audio, and PDF (output stays text-only). Developers can pick between three thinking levels (low, medium, high) to trade reasoning depth against latency and cost. Google has been shipping specialized Gemini models on a similarly fast cadence this cycle, including Gemini 3.5 Transcribe for speech-to-text work.

The model shipped alongside a second, much more restricted variant: Gemini 3.8 Flash Cyber, tuned for vulnerability research and automated patching. It sits behind Google's Fairwind Program and is limited to vetted government bodies, critical infrastructure operators, and software maintainers, so most developers will only ever touch the base 3.8 Flash model, not its security-focused sibling.

Gemini 3.8 Flash Pricing: What It Costs

Here is the part that actually answers the search intent behind gemini 3.8 flash pricing: Google did not raise the introductory rate. Gemini 3.8 Flash bills at $0.75 per million input tokens and $3.75 per million output tokens, the exact same introductory pricing Google set for 3.7 Flash, and that rate holds through December 31, 2026. Standard pricing of $1.50 / $7.50 per million tokens takes over on January 1, 2027 (Gemini API changelog). For teams weighing this against other coding assistants, see how the numbers stack up in our breakdown of Claude Code vs Cursor pricing.

That is a meaningful detail for anyone running production workloads: teams already on 3.7 Flash can move to 3.8 Flash without a line-item cost increase before the new year, then plan for the standard-rate step-up in January. The sources reviewed here do not list a separate introductory rate for Gemini 3.5 Flash, so if you are budgeting against that older model specifically, check your own billing console rather than assume parity. If you want a deeper breakdown of how usage-based LLM pricing adds up in production, our guide to Claude API pricing and cost tuning walks through the same math for Claude's models. What is confirmed is that 3.8 and 3.7 Flash share the same number, and that number is scheduled to double in four months.

On GitHub Copilot, which added Gemini 3.8 Flash to its model picker on September 3, the model bills at introductory provider pricing under Copilot's usage-based billing through December 31, 2026, mirroring the Gemini API's own cutoff date (GitHub Changelog).

How It Stacks Up on Long-Horizon Coding

Pricing only matters next to capability, and this is where 3.8 Flash's case gets more interesting than a routine Flash refresh. On Terminal-Bench 2.1, a benchmark for autonomous, terminal-based task completion, Gemini 3.8 Flash scores 90.8%, up from 3.7 Flash's 81.6%, one of the larger single-release jumps in the Flash line's recent history. On DeepSWE v1.1, a long-horizon software engineering benchmark, third-party coverage puts 3.8 Flash at 73.7% versus 65.3% for 3.7 Flash, and reports it outperforming several larger, pricier frontier models on the same tasks.

Google attributes the gain to the model taking smaller, more deliberate reasoning steps on multi-step goals, calling tools iteratively, and verifying its own work as it goes, which costs somewhat more tokens on hard tasks but catches more of its own mistakes before they compound across a long agentic run. GitHub's own early testing echoes this: it found Gemini 3.8 Flash showed "rigorous validation and persistent recovery from actionable failures" on complex terminal-based coding tasks, meaning it is more likely to notice an error mid-task and correct course rather than pile more broken steps on top of it.

The source material reviewed here does not include head-to-head benchmark numbers against Gemini 3.5 Flash, so treat any claim of a specific score gap over 3.5 Flash with caution until Google or a third party publishes one. What is documented is a clear, two-generation trend: 3.6 Flash shipped July 21, 3.7 Flash shipped August 13, and 3.8 Flash shipped roughly three weeks after that, each release posting gains on the same benchmark suite at flat pricing.

Where You Can Use It Today

Gemini 3.8 Flash is live now through the Gemini API, Google AI Studio, Antigravity, Android Studio, and the Gemini Enterprise Agent Platform, with support for caching, code execution, computer use (preview), file search, function calling, Google Maps grounding, search grounding, structured outputs, URL context, the Batch API, and both Flex and Priority inference tiers.

It is also rolling out across GitHub Copilot's model picker in Visual Studio Code, Visual Studio, the Copilot CLI, the Copilot cloud agent, the Copilot app, JetBrains IDEs, Xcode, and Eclipse, available to Pro, Pro+, Max, Business, and Enterprise subscribers. The rollout is staged, so GitHub notes that if the model has not appeared in your picker yet, check back over the following days. Business and Enterprise admins should also check their organization's model policy settings, since new models default to on unless an admin has disabled the global default or the specific model.

Gemini 3.8 Flash for Product Managers

If you are a PM evaluating which model powers your team's agentic coding workflows rather than writing the code yourself, the pricing stability is the number worth flagging to engineering: a flat rate through the end of 2026 means you can commit a workflow to 3.8 Flash now and budget the standard-rate jump for January without a mid-quarter surprise. Engineering teams already tracking spend should also look at these habits for cutting Claude Code token usage, since the same discipline applies regardless of which model sits behind the agent. The Terminal-Bench and DeepSWE gains matter less for a spec review or a competitive teardown of a rival's product, and more for anything you are asking an agent to run unattended for a while, like scaffolding a prototype from a PRD or chasing down a bug across a large codebase. If you want to build that kind of agentic workflow yourself, our Claude Code series walks through shipping code faster and building your own agents.

Gemini 3.8 Flash is a straightforward case: same introductory price as its predecessor, meaningfully higher scores on the benchmarks Google chose to publish, and a same-day rollout into GitHub Copilot that suggests the model is stable enough for production use, not a research preview. Whether it is the right default for your own long-horizon coding tasks still comes down to testing it against whatever you are running today, but the pricing question, at least, has a clear answer through the end of 2026.