GitHub Copilot adds Gemini 4 Hydrogen at $1.50/$7.50 per 1M tokens

6 min read

GitHub made Gemini 4 Hydrogen, Google's new fast Gemini 4 model, generally available in GitHub Copilot on November 13, 2026, for subscribers on the Pro, Pro+, Max, Business and Enterprise plans. The rollout is gradual, so some accounts will see the model in the picker later than others. GitHub says Gemini 4 Hydrogen is billed at the provider's list price under Copilot's usage-based billing, and Google's official Gemini API documentation lists that price at $1.50 per million input tokens and $7.50 per million output tokens, with implicitly cached input at $0.075 per million.

The addition has a fixed deadline attached. In a separate changelog entry published the same day, GitHub said Gemini 3.8 Flash will be deprecated across all Copilot experiences on December 14, 2026, and named Gemini 4 Hydrogen as its suggested replacement. In Copilot, Gemini 3.8 Flash is currently billed at $0.75 per million input tokens and $3.75 per million output tokens, a reduced rate that Google applies through December 31, 2026. Developers who move from the outgoing Flash model to Hydrogen therefore pay twice as much per uncached input token and per output token, while the cached-input rate stays at $0.075 per million.

Gemini 4 Hydrogen was launched by Google on November 12 as the successor to Gemini 3.8 Flash, with a single context window of 1,048,576 tokens shared by input, thinking and output tokens. GitHub's model pricing page did not yet carry a Gemini 4 Hydrogen row when Data Studios checked it on November 14, 2026, so the rates in this report come from Google's official model documentation, which GitHub's "provider list pricing" wording points to.

··········

GEMINI 4 HYDROGEN IN GITHUB COPILOT: PLANS AND SURFACES.

GitHub lists five paid plans for the model and a long set of clients. The free tier is absent from the announcement, which matches the general rule in GitHub's billing documentation that Copilot Free and Copilot Student users reach models only through automatic model selection and cannot pick a specific model.

........

Item	Gemini 4 Hydrogen in GitHub Copilot
Announcement	November 13, 2026, GitHub changelog
Status	Generally available, gradual rollout
Plans	Pro, Pro+, Max, Business, Enterprise
Copilot Free	Not listed (Free uses auto model selection only)
Billing	Provider list pricing, usage-based AI credits
Editors	VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse
Other surfaces	Copilot CLI, coding agent, Copilot app, github.com, GitHub Mobile, Copilot SDK
Admin control	Model policy in Copilot settings

........

The breadth of surfaces is the practical difference from many earlier model launches in Copilot, which often started in VS Code and reached other editors weeks later. Here the model is announced for the IDE extensions, the command-line tool, the autonomous coding agent and the mobile apps at once, though the gradual rollout means availability can differ from one account to the next during the first days.

For Copilot Business and Enterprise, GitHub applies its default model enablement rule. New models are switched on automatically unless an administrator has disabled the global default or turned off Gemini 4 Hydrogen specifically in the model policy. Organizations that keep the global default off must enable the model by hand before developers can select it.

··········

GEMINI 4 HYDROGEN PRICING IN COPILOT AI CREDITS.

Since Copilot moved to usage-based billing, every chat turn, agent step and CLI request is metered by token type and converted into AI credits, with one credit worth $0.01. A price of $1.50 per million tokens therefore equals 150 credits per million tokens. Code completions and next edit suggestions are excluded from this metering and remain unlimited on paid plans.

The table below compares the rate GitHub currently bills for Gemini 3.8 Flash with Google's list price for Gemini 4 Hydrogen. Prices as listed on GitHub's models and pricing page and on Google's official Gemini 4 Hydrogen documentation on November 14, 2026.

........

Model in Copilot	Input	Cached input	Output	Effective
Gemini 3.8 Flash (outgoing)	$0.75	$0.075	$3.75	Until Dec 14, 2026
Gemini 4 Hydrogen (new)	$1.50	$0.075	$7.50	From Nov 13, 2026
Change	+100%	0%	+100%

........

All figures are USD per million tokens. Google bills thinking tokens as output, and Gemini 4 Hydrogen defaults to the "high" thinking level, so the output side of a Copilot request can be larger than the visible answer. Google also states that implicit caching applies by default to prompts of at least 4,096 tokens and that explicit cache objects are not supported on this model. In Copilot, where an agent resends the same repository context on each step, the share of cached input has a large effect on the final bill.

For a wider view of what GitHub charges per plan and how it compares with Cursor, Claude Code and Codex, see our guide to AI coding agent pricing.

··········

WHAT A COPILOT AGENT SESSION COSTS ON GEMINI 4 HYDROGEN.

To make the price difference concrete, Data Studios modelled one illustrative agent session. The assumptions are 20 model calls, each sending 30,000 input tokens with 70% served from the implicit cache, and each returning 1,500 output tokens including thinking. That totals 600,000 input tokens (420,000 cached and 180,000 uncached) and 30,000 output tokens. These are assumptions for comparison, not measured Copilot traffic.

........

Model	Uncached input	Cached input	Output	Session cost	AI credits
Gemini 3.8 Flash	$0.135	$0.0315	$0.1125	$0.279	27.9
Gemini 4 Hydrogen	$0.270	$0.0315	$0.225	$0.5265	52.65
Claude Haiku 5.5	$0.018	$0.0042	$0.015	$0.0372	3.72
Claude Sonnet 5.5	$0.360	$0.042	$0.300	$0.702	70.2

........

Under these assumptions the same session costs about 89% more on Gemini 4 Hydrogen than on Gemini 3.8 Flash. The increase is a little below 100% because the cached-input rate did not change. A workload with a lower cache hit rate moves closer to a full doubling, and a workload dominated by cached repository context moves further below it. Claude Haiku 5.5 and GPT-6 Luna, both billed by GitHub at $0.10 input, $0.01 cached input and $0.50 output per million tokens, remain far cheaper per session, while Claude Sonnet 5.5 at $2.00 and $10.00 costs more than Hydrogen.

Converted into monthly allowances, the difference shows up quickly. A Copilot Pro subscription includes 1,500 AI credits (1,000 base and 500 flex), enough for about 53 sessions of this size on Gemini 3.8 Flash and about 28 on Gemini 4 Hydrogen. Pro+ includes 7,000 credits, which covers about 250 such sessions on the outgoing model and about 132 on Hydrogen. A Copilot Business seat contributes 1,900 credits to the organization's pooled allowance, roughly 68 Flash sessions or 36 Hydrogen sessions per seat, with usage beyond the pool charged at $0.01 per credit.

··········

WHAT GITHUB COPILOT USERS AND ADMINS NEED TO DO BEFORE DECEMBER 14.

Individual subscribers who rely on Gemini 3.8 Flash can select Gemini 4 Hydrogen in the model picker as soon as it appears for their account and compare results on their own repositories before the December 14 cut-off. GitHub's billing documentation says unused credits do not carry over and that allowances reset at 00:00 UTC on the first day of each calendar month. When the allowance runs out, Pro, Pro+ and Max users can upgrade and pay only the price difference, set a budget for additional usage, or wait for the reset.

Business and Enterprise administrators should check two settings. The first is whether the global default for new models is enabled, because that decides whether Gemini 4 Hydrogen and the other suggested replacements appear for developers without manual action. The second is the organization's budget for additional usage, since teams that move a high volume of agent work from Gemini 3.8 Flash to Hydrogen will consume pooled credits faster at the new rate. Our report on the six Copilot models retiring on December 14 lists the other replacements, including Claude Fable 5.2 for Claude Fable 5 and Gemini 4 Carbon for Gemini 4 Argon.

Developers who call Gemini 4 Hydrogen directly through the Gemini API, outside Copilot, face several changes documented by Google, including the removal of the temperature, top_p and top_k parameters, a new "xhigh" thinking level and an error for requests that ask for the "minimal" level. Copilot handles those parameters on the user's behalf, so they do not require action inside the editor. The model's launch details and Google's own pricing are covered in our Gemini 4 Hydrogen launch report and in the Gemini 4 Hydrogen vs Gemini 3.8 Flash comparison.

··········

GEMINI 4 HYDROGEN IN COPILOT: CONFIRMED FACTS AND OPEN POINTS.

GitHub has confirmed general availability with a gradual rollout, the five eligible paid plans, the list of supported clients, billing at provider list pricing and the admin policy behaviour. Google's documentation confirms the list price of $1.50 input, $7.50 output and $0.075 cached input per million tokens, the 1,048,576-token shared window and the absence of explicit caching. GitHub has also fixed December 14, 2026 as the end date for Gemini 3.8 Flash in Copilot.

Three points were still open on November 14. GitHub's models and pricing page had not yet added a Gemini 4 Hydrogen row, so the exact credit rates should be confirmed there once it appears. GitHub has not said whether Gemini 4 Hydrogen will join automatic model selection, which is the only way Copilot Free and Student users reach models and where Gemini 3.8 Flash currently sits. GitHub has also not published Copilot-specific quality or latency figures for the model. For a team that used Gemini 3.8 Flash as its low-cost default, the move to Gemini 4 Hydrogen roughly doubles the per-token price, and the realistic cost increase for agent sessions with good cache reuse sits somewhat below that figure.

··········

·····

DATA STUDIOS

·····

datastudios.org