Coding-agent budgets now have two frontier options a day apart: Google opened Gemini 4 Carbon to paying developers on Nov. 10, one day after Anthropic shipped Claude Fable 5.2. Both target long-horizon coding, not everyday chat, so access, safeguards, orchestration, and cost per task matter more than the launch. Google calls the $2 and $10 token rates an introductory price, so the sticker is not a lock-in.
Updated November 10, 2026
Key Takeaways
- Carbon opened Nov. 10, a day after Fable 5.2, for long-horizon coding.
- List rates are $2/$10 per million for Carbon and $8/$40 for Fable.
- Fable is open to all; Enterprise Claude Security also includes Mythos.
- Google calls Carbon’s rate introductory; swarms burn quota faster.
Who Can Actually Run Each Model?
Google’s Carbon availability post opens the API in Google AI Studio and Gemini for business to paying developers and enterprises. It is also rolling out to Google AI Ultra subscribers in Antigravity, the Gemini app, and AI Mode in Search. The Antigravity blog says scientists and researchers can apply for a chance at Argon and Carbon on the highest usage tier free for 6 months.
Anthropic’s announcement splits one model. Fable 5.2 is available to all users because of cybersecurity and biology safeguards. Organizations can apply for Mythos through the Life Sciences and Cyber Verification Programs, and Claude Enterprise customers can use Claude Security, which includes Mythos 5.2, even if they are not program members. For most coding, Anthropic still points users to Opus 5.5.
Analysis: Carbon’s general API is paywalled and Fable’s is broadly open. Mythos is not limited to the verification programs.
How Do Safeguards Change the Work?
Anthropic says Fable 5.2’s safeguards will block most cybersecurity-related tasks, with its strongest-ever protections against distillation attacks. On CVE Rediscovery, Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1. That cyber path is the verification programs or Claude Security for Enterprise customers.
Google released Carbon to a set of cybersecurity defenders via Fairwind two weeks before the public opening. Google says that Fairwind access leads on SEC-bench Pro, ExploitGym, and CWE-bench v1. Analysis: Fable will often refuse the cyber step; the benchmark lead Google cites is that Fairwind access.
How Does Agent Orchestration Differ?
Google is previewing /swarm in the Antigravity app and CLI: a coordinator, workers in isolated git worktrees, and adversarial verifiers on a shared board. Idle agents park at zero token cost. Carbon is for coordinator, architecture, cybersecurity auditing, and adversarial verification; lighter models edit files. Antigravity warns that parallel multi-agent systems consume quota much faster than a standard chat.
Fable 5.2 can advise Haiku, Sonnet, or Opus, or delegate simpler tasks to smaller models. Analysis: Google is shipping the harness; Anthropic is defining a role inside yours. Fan-out, not the chat price, moves the bill, and review load still counts, as our coding-agent task gap reporting shows.
What Do Coding-Agent Budgets Look Like Beyond List Price?
Analysis of the published rates: Fable 5.2 lists at four times Carbon on input and output.
| Factor | Gemini 4 Carbon | Claude Fable 5.2 |
|---|---|---|
| Input list price | $2 per million tokens | $8 per million tokens |
| Output list price | $10 per million tokens | $40 per million tokens |
| General access | Paying developers and AI Ultra | All users |
| Cyber-capable path | Fairwind Program | Mythos via programs or Claude Security |
| Orchestration | /swarm preview in Antigravity | Advisor or Orchestrator |
Fable fast mode is $16 and $80 per million tokens, up to 2.5 times faster. AI Studio lists implicit caching only, not explicit caching. Google says Carbon ranks first on CursorBench 4.0, Perplexity’s WANDR, and Zapier’s AutomationBench at more than 2X lower cost per task than the next best external model.
Anthropic says High-effort Fable 5.2 scored about level with Opus 5.5 at Max effort on FrontierCode 2.0, for roughly half the cost per task. On Medium effort, the default in Claude apps and Claude Code, Fable 5.2 can accomplish challenging tasks with fewer tokens than Fable 5.1.
Analysis: Google also says Carbon uses more tokens than Argon to explore, reason, and run extended verification loops, and that swarms burn quota faster despite zero-token idle time. Anthropic’s savings are versus Opus and Fable 5.1, not versus Carbon. Measure tokens per merged change, keep a routing cut such as our TokenEconomizer notes in the plan, and re-check the introductory rate before a long default.
- Pin Carbon or Fable to architecture and hard debug; leave routine edits on Argon or Opus.
- Budget fan-out separately. Zero-token idle time does not cap parallel spend.
- Send cyber work to Fairwind or Mythos, including Enterprise Claude Security. Do not expect Fable safeguards to allow it.
- Re-check Carbon’s introductory rate before you lock a default.
Frequently asked questions
When did Carbon and Fable 5.2 become available?
Google made Gemini 4 Carbon available to paying developers on Nov. 10, 2026, a day after Anthropic introduced Fable 5.2 and Mythos 5.2.
How do the token prices compare?
Carbon is listed at $2 per million input tokens and $10 per million output tokens. Fable 5.2 is $8 and $40, or $16 and $80 in fast mode.
Can Fable 5.2 handle cybersecurity agent work?
Anthropic says Fable 5.2’s safeguards will block most cybersecurity-related tasks. Mythos 5.2 is available through verification programs, and Enterprise customers can use it in Claude Security without being program members.
Does a lower list price mean a cheaper agent bill?
Not by itself. Google says Carbon uses more tokens than Argon and that swarms consume quota much faster than chat, on different cost-per-task tests than Anthropic’s.
