Fable 5.2 and Gemini 4 Carbon: what developers get

Two frontier models for coding shipped on consecutive days. On Monday 9 November, Anthropic released Claude Fable 5.2 and cut its price by 20%. On Tuesday 10 November, Google made Gemini 4 Carbon available to paying API users at an introductory price that doubles at the end of the month. Both companies say their new top model beats the other's on agentic coding, and both tell you to keep using their cheaper mid-tier model for most work.

Here is what each release is, what it costs, what the benchmarks do and do not show, how to try each one from your terminal, and when it makes sense to stay on Opus 5.5 or Gemini 4 Argon.

What shipped

Claude Fable 5.2 and Mythos 5.2 (9 November)

Fable and Mythos are the same underlying model. Fable ships with cyber and biology safeguards and is available to everyone on a paid plan. Mythos is restricted to organisations in Anthropic's Cyber and Life Sciences Verification Programs, plus Claude Enterprise customers through Claude Security. For most developers, Fable is the one that matters.

Anthropic's launch post makes three claims a developer would care about:

• Faster and cheaper. Output is up to 30% faster than Fable 5.1, and the price fell from $10 and $50 per million input and output tokens to $8 and $40.
• Better on agentic coding. Its highest score on the FrontierCode 2.0 benchmark is 49.1% at the Xhigh effort level, which Anthropic says is the highest of any released model it tested.
• Not the default. Anthropic still calls Opus 5.5 its "strongest daily driver" and recommends moving up to Fable only when a task needs it. Dianne Penn, its head of product management for research, told CNBC that Fable 5.2 is primarily for "power users or power use cases."

The API model ID is claude-fable-5-2. It is available on the Claude API, in the Claude apps and Claude Code, and through Google Cloud, Microsoft Foundry and the Claude Platform on AWS. The Free plan does not include it.

Gemini 4 Carbon (10 November)

Google released Gemini 4 Carbon to the paid tier of the Gemini API in AI Studio and Android Studio, to enterprises through Gemini for business and Antigravity, and to Google AI Ultra subscribers in the Gemini app. Its developer-facing claims:

• Top of three agent benchmarks. Google says Carbon ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench, at more than twice lower cost per task than the next best external model.
• A new /swarm mode in Antigravity, in preview, that turns one prompt into a coordinated team of agents, with Carbon handling architecture and auditing while faster models do parallel work. Google's pitch is the same multi-agent idea that Meta, OpenAI and GitHub all bought into last week.
• Also not the default. Carbon uses more tokens than Argon for deeper reasoning and verification, so Google positions Gemini 4 Argon as the model for most software engineering work.

Prices side by side

All prices are per million tokens, from the vendors' own pricing pages, checked on 11 November 2026.

Model	Input	Output	Notes
Claude Fable 5.2	$8	$40	Cache reads $0.20. Fast mode $16 / $80
Claude Fable 5.1	$10	$50	Previous generation
Claude Opus 5.5	$4	$20	Anthropic's recommended default
Claude Sonnet 5.5	$2	$10
Gemini 4 Carbon (until 30 Nov)	$2	$10	Introductory, same as Argon
Gemini 4 Carbon (from 30 Nov)	$4	$20

Two things stand out. First, Fable 5.2 is still twice the price of Opus 5.5 and four times Sonnet 5.5, so the 20% cut is a smaller deal than the headline suggests unless you were already paying for Fable. Second, Carbon's introductory price is a three-week promotion, and at its real price it matches Opus 5.5 exactly. Do not build a cost model on the $2 figure.

What the benchmarks say, and what they do not

Anthropic published a comparison table in its launch post. These are the rows that matter for coding, with the competitor numbers as Anthropic reports them:

Benchmark	Fable 5.2	Opus 5.5	GPT-6 Astra	Gemini 4 Carbon
FrontierCode 2.0 (agentic coding)	49.1% (Xhigh)	45.5%	42.8%	41.1% (Medium)
FrontierSWE v2 (agentic coding)	69.9%	62.3%	65.5%	61.2%
Terminal-Bench-Science 0.1	74.6%	58.7%	68.1%	64.8%
AutomationBench (business workflows)	51.4%	42.5%	41.4%	53.9%

Read these with the usual care:

• They are vendor-run. Anthropic ran the competitor models. Note the effort levels in the first row: Fable at Xhigh, Carbon at Medium. Google's own post claims first place on AutomationBench and WANDR, and Anthropic's table agrees on AutomationBench, which is a small reason to trust the rest.
• The effort level is half the story. Anthropic says Fable 5.2 at High effort scores 45.3% on FrontierCode, within a rounding error of Opus 5.5, at about half the cost per task. The Xhigh number is the headline, but the High number is the one that changes a bill.
• One independent result backs the trend. The Remote Labor Index, run by Scale and the Center for AI Safety, asks agents to complete 240 real paid Upwork projects, graded by human experts who open the files in the real software. Fable 5.2 completed 41.3% of them. Opus 5.5 scored 21.25%, GPT-6 Astra 20.83%, and the best system a year ago managed 2.5%. The tasks are freelance jobs such as CAD, video editing and data analysis rather than pull requests, but it is the clearest non-vendor signal that the jump from 5.1 is real.

Anthropic also says over 90% of changes merged into its own core codebase are now written by Claude with human guidance, and that one engineer had Fable 5.2 build a prototype operating system over eight days with more than 1,000 subagents. Treat those as marketing stories, not benchmarks.

Try Fable 5.2 from Claude Code

No Fable model is the default on any plan, so you select it explicitly. Fable 5.2 needs Claude Code 2.1.321 or later.

/model fable        # switch inside a session; Enter saves it as your default, s is session-only
/effort high        # medium is the default for Fable 5.2; xhigh and max exist
/fast               # optional: up to 2.5x faster output at double the token price

Or from the shell:

claude --model fable --effort high

Three billing details before you leave it running overnight:

1. Plan limits. Anthropic's launch post says Fable is included for up to 50% of weekly usage limits on Pro and standard Team seats, with one to three "Fable resets" depending on subscription length. The Claude Code docs add that Fable usage may bill to usage credits rather than your plan's included limits. Run /usage before and after a session.
2. Fast mode is usage credits only on Pro, Max, Ultra, Team and Enterprise. It is never covered by the subscription, and the first time you turn it on mid-conversation you pay the fast-mode uncached price for the whole context. Turn it on at the start of a session or not at all.
3. Safety fallbacks. Requests flagged by the cyber or biology classifiers fall back to an Opus model automatically. Anthropic says this now happens about 55% less often than on Fable 5.1, but if you do security work, expect some.

If you have not used Claude Code at all, the Claude Code profile covers plans and setup.

Try Gemini 4 Carbon

For API access, Carbon is on the paid tier of the Gemini API through Google AI Studio, and inside Android Studio. If you pay for Google AI Ultra, it is also in the Gemini app, AI Studio and Antigravity, where the /swarm preview lives. Google's post does not mention Gemini CLI, so assume terminal access goes through the API for now.

If you want to compare it against Fable on your own repository, the cheapest honest way is the same task through both APIs at the price each will actually charge in December: $4 and $20 for Carbon, $8 and $40 for Fable. A test at the introductory price tells you nothing about next month's bill.

When to stay on Opus 5.5 or Argon

Both vendors say the same thing in their own launch posts: the new top model is for the hardest work, and the mid-tier model is the default. That is good advice, and the economics explain why.

• Daily edits, tests, refactors in files you understand: Opus 5.5 or Argon. Fable at Medium effort is Anthropic's own default, and at that level it is not far from Opus on coding benchmarks while costing double per token.
• Long autonomous sessions on ambiguous problems: this is what Fable is for. Multi-package migrations, a root-cause hunt across subsystems, a redesign where the agent has to hold a lot of the codebase at once. Google describes Carbon's strengths in almost the same words.
• Anything where review time is your bottleneck: a more capable model produces more code to review, not less. The multi-agent post made the argument last week and nothing here changes it.

A practical pattern is to keep opus as your default and switch with /model fable for the one task a day that deserves it, then switch back. Claude Code's picker makes a session-only switch a single keypress, so there is no reason to leave the expensive model on.

The short version

• Fable 5.2 is faster, 20% cheaper, and leads Anthropic's own coding benchmarks and the independent Remote Labor Index. It is still twice the price of Opus 5.5 and is not the default anywhere.
• Gemini 4 Carbon matches Opus 5.5's price from 30 November and is on a half-price promotion until then. Google, like Anthropic, tells you to use its mid-tier model for everyday coding.
• To try Fable today: update Claude Code to 2.1.321 or later, run /model fable, and watch /usage. Fast mode costs double and comes out of usage credits.
• To try Carbon today: AI Studio's paid tier or Android Studio. Price your tests at the post-November rate.

Frequently asked questions

How much does Claude Fable 5.2 cost?

On the Claude API, $8 per million input tokens and $40 per million output tokens, 20% below Fable 5.1. Cache reads are $0.20 per million. Fast mode, which generates tokens up to 2.5 times faster, is $16 and $80. Checked on Anthropic's pricing page on 11 November 2026.

How do I use Fable 5.2 in Claude Code?

Run /model fable inside a session, or start one with claude --model fable. It needs Claude Code 2.1.321 or later. Fable is not the default on any plan, and usage may bill to usage credits rather than your plan's included limits, so check /usage before a long session.

Is Gemini 4 Carbon free to try?

Not for developers. It is on the paid tier of the Gemini API in Google AI Studio and in Android Studio, and Google AI Ultra subscribers get it in the Gemini app and Antigravity. Introductory API pricing is $2 and $10 per million tokens until 30 November 2026, then $4 and $20.