Gemini 4 Carbon costs $2 and $10 per million input and output tokens until November 30, 2026, then $4 and $20: the same as Claude Opus 5.5 and half of Claude Fable 5.2. No independent test compares the three on production issues yet. Each vendor leads its own table, so try Carbon on one real issue in Copilot.
Google made Gemini 4 Carbon available on November 10, 2026, and GitHub Copilot added it the same day.1,3 We checked every price, plan rule and score below on Google's, GitHub's and Anthropic's own pages on November 16, 2026, and cite them inline. We have not run our own benchmark, and we say where a number comes from the vendor it favors. Fixter (fixter.dev) is our product; it comes in where the issue comes from, not in the model choice.
Gemini 4 Carbon, Opus 5.5 and Fable 5.2 side by side
Google calls Carbon its strongest model, built for complex, multi-step coding and debugging, and says it spends more tokens than Gemini 4 Argon to reason and verify.1 Here it is next to the two Claude models most people debug with today.
| Gemini 4 Carbon | Claude Opus 5.5 | Claude Fable 5.2 | |
|---|---|---|---|
| Maker | Anthropic | Anthropic | |
| API price per million tokens | $2 input, $10 output until November 30, 2026; then $4 and $20 | $4 input, $20 output | $8 input, $40 output |
| Cached input per million tokens | $0.10 until November 30, 2026; then $0.20 | $0.20 | $0.20 |
| GitHub Copilot plans | Pro+, Max, Business, Enterprise; gradual rollout | Pro+ and Max among the individual plans | Pro+, Max, Business, Enterprise; admins turn the policy on |
| Billing in Copilot | Usage-based, at the API price | Usage-based, at the API price | Usage-based, at the API price |
| FrontierCode 2.0, Anthropic's table | 41.1% at medium effort | 45.5% | 49.1% at xhigh effort |
| What its maker says | Google's strongest model; Argon stays its pick for most software engineering | Anthropic's recommended starting point | For the hardest open-ended technical work |
| Where you can run it | Copilot, Gemini API in Google AI Studio, Antigravity and the Gemini app on Google AI Ultra | Claude Code, Copilot, Claude API | Claude Code 2.1.321 or later, Copilot, Claude API |
Gemini prices from Google's launch post and the Gemini API pricing page; Claude prices from Claude's pricing docs; Copilot plans and billing from GitHub's changelog and docs; FrontierCode from Anthropic's launch post, on its own evaluation; the Claude Code version from its changelog. All checked November 16, 2026.1,2,3,4,5,6,7,8,17
Two things stand out. Until the end of November, Carbon costs half as much per token as Opus 5.5 and a quarter of Fable 5.2. From December 1, 2026 it costs the same as Opus 5.5.2 And both vendors tell you to start one model down: Google says Argon "remains the ideal choice for the majority of software engineering and knowledge work", and Anthropic recommends Opus 5.5 before Fable.1,5 In GitHub Copilot, Google's half of that advice has an end date: Copilot removes Gemini 4 Argon on December 14, 2026 and names Carbon as its replacement.19 After that, the Gemini choice in Copilot is Carbon or Hydrogen.
What the benchmarks say, and who ran them
There are two sets of numbers, and each comes from a vendor's own launch post.
- Anthropic's table. FrontierCode 2.0 measures whether a model's code changes would be merged by the maintainers of open source repositories. Anthropic's Fable 5.2 post lists Fable 5.2 at 49.1% at xhigh effort, Opus 5.5 at 45.5% and Gemini 4 Carbon at 41.1%.5 Carbon's score is labelled medium effort, and the post does not say why or how the number was obtained. Carbon had only been released to a group of security teams when that post came out.1
- Google's claims. Google reports Carbon first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench, at less than half the cost per task of the next best external model.1 The post gives ranks, not scores, and does not name the models it beat.
So the honest reading is narrow. On the one coding table that has all three, Carbon trails both Claude models, at a lower effort setting. On the benchmarks Google picked, Carbon leads and costs less per task. None of these is a production debugging test, where the model reads logs and traces and has to find a cause it was not handed. Treat both sets as a reason to try, not a result.
Per token prices also understate the difference between a small model and a large one. Google says Carbon uses more tokens than Argon on the same work, and Anthropic reports Fable 5.2 at high effort about level with Opus 5.5's best score for roughly half the cost per task.1,5 What you pay for an issue is tokens times price, and the first number moves more than the second.
Which one for a production issue
Most production issues are small once you can see them. The stack trace points into your code, or the deploy diff does, and any of the three fixes it. The model starts to matter when the reasoning is the hard part. This is our reading of the facts above, not a measurement.
Try Gemini 4 Carbon when:
- You already work in Copilot. It is one more row in the model picker, with no new tool to set up.3 If you debug with Gemini 4 Argon there today, plan the switch now: GitHub deprecates Argon across Copilot on December 14, 2026 and suggests Carbon in its place.19
- The issue needs a lot of reading. Searching logs and walking traces is many tool calls. Until November 30, 2026 Carbon does that at half the per token price of Opus 5.5.2,6
- Claude already tried. A second attempt on the same model tends to repeat the first. A model from another lab starts from different habits.
Stay on Claude when:
- You work in Claude Code. Carbon is not in it. There the choice is between Opus 5.5 and Fable 5.2, which we cover in Claude Fable 5.2 vs Opus 5.5 in Claude Code.
- The cause crosses services and you want the top score. On the one shared coding table, Fable 5.2 leads.5 It also costs twice to four times as much per token.
- Your Copilot plan does not have Carbon yet. It is not on Copilot Free or Pro, the rollout is gradual, and on Business and Enterprise an admin can switch it off.3,8
Where Gemini 4 Hydrogen fits
Google released Gemini 4 Hydrogen on November 12, 2026 as the fast model next to Carbon and Argon. Google aims it at everyday coding, including issue resolution and debugging, and at quick jobs like running tests and verifying a change. It runs alone or as the worker next to Carbon.15
- Price. $1.50 per million input tokens and $7.50 per million output tokens on the Gemini API, with a 1M token context window. Unlike Carbon, it has a free tier.2,15
- Where it runs. The Gemini API in Google AI Studio, Android Studio, the Gemini app, and Antigravity for Google AI Pro and Ultra subscribers.15 GitHub added it to Copilot on November 13, 2026 for Pro, Pro+, Max, Business and Enterprise, billed at the API price, with a gradual rollout.4,18 So it is on Copilot Pro, where Carbon is not.8 The Gemini CLI model docs do not mention it.13
For a production issue, reach for Hydrogen when the stack trace or the deploy diff already points at the cause and the work is reading logs, applying the fix and checking it. Keep Carbon for the issue where nobody can see the cause yet. This is our reading of Google's own positioning, not a test we ran.
How to run Carbon on your production data from Copilot
The model only reasons about what it can see. Connect your monitoring tool over MCP, so Copilot reads the real error and not your summary of it. These steps use Fixter (fixter.dev) in VS Code; any monitoring tool with an MCP server works the same way.
- Check your plan. Carbon is on Copilot Pro+, Max, Business and Enterprise, and is billed at the provider's list price under usage-based billing. The rollout is gradual, so it may not be in your picker yet. On Business and Enterprise, admins control it through the model policy in Copilot settings.3
- Create a Fixter API key. In Fixter, open Settings, then API keys. The value is shown once. Put it in an environment variable named
FIXTER_API_KEYon the machine that runs VS Code, not in a file you commit.10 - Add the server to Copilot. Add the entry below to
~/.copilot/mcp-config.json, your personal Copilot MCP configuration, under the top-levelmcpServerskey. Thehttptype is a remote server, and header values can reference environment variables in the braced form.9,16 - Pick Gemini 4 Carbon in the Copilot model picker.3
- Ask for the open issues. Ask Copilot for the open issues in the area that broke. It opens the issue with its stack frames and trace and, on Fixter's paid plans, the root cause and who is affected.11
- Review the fix. Log lines and error messages can carry text your users typed, so tell the model to treat them as data, and review the change like any other before it merges.
An illustration of the steps above, not a recorded session.
The MCP connection does not change when you switch models: the same Fixter tools answer Carbon, Opus 5.5 or Fable 5.2. To run this on a schedule instead of by hand, see Copilot Automations with an MCP server. Fixter finds and fixes the issues in your app in production, from your coding agent: errors, failing infrastructure, slow requests and downtime. It finds the root cause and fixes it: in your agent over MCP, or as a pull request ready for your review. Nothing to learn, built for small teams. Free plan, paid from €20 a month.12
What about Gemini CLI and Antigravity
Google has not said Carbon or Hydrogen runs in Gemini CLI. On November 16, 2026 the Gemini CLI model docs list Gemini 3 and Gemini 2.5 models, say the CLI has not been optimized for Gemini 4 Argon and point to Antigravity for it.13 The 0.68.0 release notes from November 10 do not mention Carbon.14
Google's own route is Antigravity, where Carbon is rolling out to Google AI Ultra subscribers, and the Gemini API in Google AI Studio for paying developers. Carbon is not on the API's free tier.1,2 If you are not on Ultra, Copilot is the shortest way to put Carbon next to your code and your production data today.
Key takeaways
- Carbon is $2 and $10 per million tokens until November 30, 2026, then $4 and $20, the same as Opus 5.5
- Anthropic's table has Carbon behind both Claude models; Google's has it first. Neither tests production debugging
- In Copilot, Carbon is on Pro+, Max, Business and Enterprise, billed at the API price
- Carbon is not in Claude Code, and Google has not said it runs in Gemini CLI
- Gemini 4 Hydrogen is the fast sibling at $1.50 and $7.50, for issues whose cause is already visible; in Copilot it is on Pro and up
- Whichever model you pick, get the production error into the session first
Frequently asked questions
Is Gemini 4 Carbon better than Claude Fable 5.2 for coding?
It depends on whose table you read. On Anthropic's FrontierCode 2.0 table, Carbon scores 41.1% at medium effort, against 49.1% for Fable 5.2 at xhigh effort and 45.5% for Opus 5.5. Google reports Carbon first on CursorBench 4.0, at less than half the cost per task of the next best external model. No independent test compares them on production issues yet.
How much does Gemini 4 Carbon cost?
$2 per million input tokens and $10 per million output tokens on the Gemini API until November 30, 2026, then $4 and $20. That later price equals Claude Opus 5.5 and is half of Claude Fable 5.2 at $8 and $40. Carbon is not available on the API's free tier.
Is Gemini 4 Carbon in GitHub Copilot?
Yes, since November 10, 2026, for Copilot Pro+, Max, Business and Enterprise. The rollout is gradual, and Business and Enterprise admins control it through the model policy. GitHub bills it at the provider's list price under usage-based billing.
Can I use Gemini 4 Carbon in Gemini CLI?
Google has not said so. On November 16, 2026 the Gemini CLI model docs list Gemini 3 and Gemini 2.5 models and send Gemini 4 Argon users to Antigravity, and the 0.68.0 release notes do not mention Carbon. Google's launch post names Antigravity, the Gemini app and Google AI Studio.
What is Gemini 4 Hydrogen?
Google's fast Gemini 4 model, released November 12, 2026, for everyday coding and agentic work, including debugging. It costs $1.50 per million input tokens and $7.50 per million output tokens on the Gemini API, with a 1M token context window and a free tier. GitHub added it to Copilot on November 13, 2026 for Pro, Pro+, Max, Business and Enterprise, so unlike Carbon it is on Copilot Pro. The Gemini CLI docs do not mention it.
How do I give Gemini 4 Carbon access to my production logs?
Use it in a client that speaks MCP and connect your monitoring tool's MCP server. In GitHub Copilot, add the server to your Copilot MCP configuration, pick Gemini 4 Carbon in the model picker and ask for the open issues. For Fixter (fixter.dev) the server is https://mcp.fixter.dev, with an API key in the Authorization header.
Should I switch from Claude to Gemini 4 Carbon for debugging?
Not on the benchmarks alone. If you work in Copilot, trying Carbon on one real issue is a model picker change, and until November 30, 2026 it costs half as much per token as Opus 5.5. If you work in Claude Code, Carbon is not in it, so the choice there stays between Opus 5.5 and Fable 5.2.
Sources (19), all checked November 16, 2026
- Google, Gemini 4 Carbon is now available (November 10, 2026): positioning, introductory price, where it is available, benchmark claims
- Google AI Studio, Gemini API pricing: Gemini 4 Carbon rates until November 30, 2026 and from December 1, 2026, no free tier
- GitHub changelog, Gemini 4 Carbon in GitHub Copilot (November 10, 2026): plans, billing, rollout, where to select it
- GitHub docs, Models and pricing for GitHub Copilot: per-token prices for Gemini 4 Carbon, Claude Opus 5.5 and Claude Fable 5.2
- Anthropic, Claude Fable 5.2 and Mythos 5.2 (November 9, 2026): FrontierCode 2.0 table, which model to start with
- Claude docs, pricing: Claude Fable 5.2 and Claude Opus 5.5 per million tokens
- GitHub changelog, Claude Fable 5.2 in GitHub Copilot (November 9, 2026)
- GitHub docs, Plans for GitHub Copilot: models on Pro, Pro+ and Max
- VS Code docs, MCP configuration reference: Copilot configuration files, headers, environment variable references
- Fixter docs, MCP server: endpoint, API key header
- Fixter docs, investigations: root cause, who is affected, pick-up in your coding agent
- Fixter pricing
- Gemini CLI docs, model selection
- Gemini CLI releases: 0.68.0 (November 10, 2026)
- Google, Introducing Gemini 4 Hydrogen (November 12, 2026): what it is for, price, context window, where it is available
- GitHub docs, Adding MCP servers for GitHub Copilot CLI: ~/.copilot/mcp-config.json, mcpServers, remote HTTP servers with headers
- Claude Code changelog, 2.1.321: Claude Fable 5.2 added as the default Fable model
- GitHub changelog, Gemini 4 Hydrogen in GitHub Copilot (November 13, 2026): plans, billing, rollout, where to select it
- GitHub changelog, Upcoming deprecation of selected GitHub Copilot models in December (November 13, 2026): Gemini 4 Argon and Claude Fable 5 deprecated on December 14, 2026, with suggested replacements