> ## Content Index
> Fetch the complete content index at: https://claudefox.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Gemini 4 Carbon vs Claude Opus 5.5 and Fable 5.2: benchmarks and pricing compared
- URL: https://claudefox.com/gemini-4-carbon-vs-claude-opus-5-5-fable-5-2/
- Published: 2026-11-18T04:15:41.000Z
- Updated: 2026-11-18T04:16:59.000Z
- Description: Gemini 4 Carbon costs half of Claude Opus 5.5's per-token price until November 30, then exactly the same. Google's Carbon table and Anthropic's Fable 5.2 table each include the other's models, and they disagree on several rows. Here are the published numbers, with sources and dates.
- Author: hupicake
- Tags: Research

Google made Gemini 4 Carbon generally available on November 10, two weeks after first giving it to cybersecurity defenders in its Fairwind Program. Google calls it its strongest model. On the Claude side, the matching tier is Claude Opus 5.5 (released September 22) and, above it, Claude Fable 5.2 (released November 9). Google's Carbon benchmark table includes Claude models, and Anthropic's Fable 5.2 table includes Carbon, so both sides have put numbers on the record.

We have not tested these models ourselves. This is a comparison of published figures: Google's and Anthropic's tables, both price pages, and Artificial Analysis, an independent benchmarking site. Each number names its source. All were read on November 18, 2026.

## How much does Gemini 4 Carbon cost compared with Opus 5.5 and Fable 5.2?

Carbon is half of Opus 5.5's per-token price until November 30 and the same price from December 1; Fable 5.2 costs twice Opus 5.5's price, and four times Carbon's until November 30.

- Gemini 4 Carbon (Gemini API, paid tier only): $2 input and $10 output per million tokens through November 30, 2026; $4 and $20 from December 1\. Context caching goes from $0.10 to $0.20 per million. Google's footnote says the December price applies to the Carbon model intended for public access. There is no free tier for Carbon.
- Gemini 4 Argon follows the same schedule: $2 and $10 now, $4 and $20 from December 1.
- Claude Opus 5.5: $4 input and $20 output per million tokens; cache hits $0.20, 5-minute cache writes $5\. Fast mode, up to 2.5 times faster, costs $8 and $40.
- Claude Fable 5.2: $8 input and $40 output per million tokens, down from $10 and $50 for Fable 5.1; cache hits $0.20, 5-minute cache writes $10\. Fast mode costs $16 and $80.

For a task that reads 200,000 tokens and writes 20,000, before caching: Carbon costs $0.60 until November 30 and $1.20 after; Opus 5.5 costs $1.20; Fable 5.2 costs $2.40\. Per-token price is only half the story, because these models spend very different numbers of tokens on the same work, as the next section shows.

## What does each model cost per task in independent tests?

On Artificial Analysis's index, Carbon at today's launch price cost about half as much per task as Opus 5.5 for a one-point-higher score; at December prices the two would cost about the same.

- Claude Fable 5.2 (max effort): index score 65, ranked 1st of 92; $26.54 per task; 40.7 minutes per task.
- Gemini 4 Carbon (high thinking): score 60, ranked 7th; $5.48 per task at the $2/$10 launch price; 33.2 minutes per task.
- Claude Opus 5.5 (max effort): score 59, ranked 8th; $10.04 per task; 24.5 minutes per task.
- Gemini 4 Argon (high thinking): score 50; $4.97 per task; 29.1 minutes per task.
- Output tokens across the index: Fable 5.2 275 million, Opus 5.5 218 million, Carbon 205 million, Argon 113 million. Input tokens: Fable 5.2 28.3 billion, Opus 5.5 15.6 billion, Carbon 7.62 billion.

Our estimate, not an Artificial Analysis figure: because every Carbon token price doubles on December 1, the same runs would cost about $10.96 per task at the new prices, slightly more than Opus 5.5's $10.04\. Artificial Analysis also lists output speed: 62 tokens per second for Carbon, 118 for Opus 5.5 and 112 for Fable 5.2\. Its time to first answer token was 73 seconds for Carbon, 215 for Fable 5.2 and 355 for Opus 5.5 at these high-effort settings. That is not chat latency: per index task, Carbon still took longer than Opus 5.5 (33.2 against 24.5 minutes).

## Which model wins on coding benchmarks?

No single winner: Carbon leads Google's Terminal-Bench 4.0 run, Opus 5.5 leads FrontierSWE between the two, and Fable 5.2 leads every coding row in Anthropic's table.

Google's Carbon table on deepmind.google compares Carbon with Gemini 4 Argon, GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1\. Its Fable column is Fable 5.1, not Fable 5.2.

- Terminal-Bench 4.0, run by Artificial Analysis: Carbon 65.2%, Opus 5.5 59.6%, Fable 5.1 52.0%.
- FrontierSWE 2: Opus 5.5 62.3%, Carbon 61.2%, Fable 5.1 56.3% (GPT-6 Astra tops the row at 65.5%).
- SWE-sweep: Carbon 9.6%, Opus 5.5 3.6%, Fable 5.1 2.9%.
- OSWorld-2.1, offline subset (computer use): Opus 5.5 87.2%, Carbon 81.5%, Fable 5.1 76.8%.
- AutomationBench (Zapier): Carbon 53.9%, Opus 5.5 42.5%, Fable 5.1 31.4%.
- GraphWalks long context, 256k to 1M tokens: Carbon 96.2%, Opus 5.5 66.8%, Fable 5.1 65.0%.

Anthropic's Fable 5.2 table compares Fable 5.2, Fable 5.1, Opus 5.5, GPT-6 Astra and Gemini 4 Carbon:

- FrontierSWE v2: Fable 5.2 69.9%, Opus 5.5 62.3%, Carbon 61.2%.
- FrontierCode 2.0: Fable 5.2 49.1% (xhigh effort), Opus 5.5 45.5%, Carbon 41.1%. Anthropic marks Carbon's score as medium effort, not its highest setting.
- OSWorld 2.1 (offline set, partial): Fable 5.2 89.3%, Opus 5.5 87.2%, Carbon 81.5%.
- AutomationBench: Carbon 53.9%, Fable 5.2 51.4%, Opus 5.5 42.5%. Anthropic's own table puts Carbon ahead of Fable 5.2 on this row.
- Terminal-Bench-Science 0.1: Fable 5.2 74.6%, Carbon 64.8%, Opus 5.5 58.7%.

## Where do Google's and Anthropic's numbers disagree?

They agree on FrontierSWE, AutomationBench and OSWorld for Opus 5.5, but report different scores for the same Claude models on Terminal-Bench, Terminal-Bench-Science and Chartography.

- Terminal-Bench 4.0, Opus 5.5: Anthropic reports 66.4% at xhigh effort in its own setup; it says that setup reproduces the public leaderboard's Opus 5 score within noise. Google's table, run by Artificial Analysis, shows 59.6%. Carbon's 65.2% comes from the Artificial Analysis run, so the like-for-like comparison is 65.2% against 59.6%.
- Terminal-Bench-Science 0.1, Fable 5.1: Google lists 43.3%, Anthropic 52.6%. Opus 5.5 is close in both (59.0% and 58.7%).
- OSWorld 2.1, Fable 5.1: Google lists 76.8%, Anthropic 83.5%.
- AutomationBench, Opus 5.5: 42.5% in both Google's table and Anthropic's Fable 5.2 page, but 40.0% on Anthropic's Opus 5.5 page. Anthropic's footnote says Zapier ran that earlier test without fallback models, so safeguard interventions counted as failures.
- Chartography: Google lists Opus 5.5 at 66.3% and Fable 5.1 at 46.2%; Anthropic lists 64.4% and 44.8% (no tools). Carbon is 72.5% in both.
- Artificial Analysis Index: Google's table uses version 5.0 (Carbon 59, Opus 5.5 58); the live site, on version 5.0.1, shows 60 and 59.
- FrontierSWE matches exactly in both tables for Carbon, Opus 5.5, Fable 5.1 and GPT-6 Astra, the cleanest row to compare.

The pattern is common in vendor tables: each lab runs its own models at their best setting and harness, and takes competitor numbers from wherever it can. Where a row was run by a third party for every model, as with FrontierSWE, the numbers line up.

## What about context, output limits and thinking settings?

All three read up to 1 million tokens; Carbon has three thinking levels, while Opus 5.5 and Fable 5.2 use adaptive thinking with adjustable effort.

- Gemini 4 Carbon: thinking levels low, medium (default) and high. The default synchronous output cap is 65,536 tokens when streaming and 32,768 when not. Longer outputs, up to about 1 million tokens shared with the input, need Google's long-decode mode (a background request or client-side continuation).
- Claude Opus 5.5: 1M context, 128K max output, adaptive thinking that is always on, default effort medium.
- Claude Fable 5.2: 1M context, 128K max output, adaptive thinking that is always on, default effort high. Fable models require 30-day data retention for safety monitoring by default; Anthropic says Enterprise Frontier Safeguards make it compatible with zero data retention, and eligible customers can keep using zero retention until those are available.

## Which should you choose for agentic coding?

If you are on Claude, Opus 5.5 stays the default and Fable 5.2 is for the hardest work; Carbon is worth an eval mainly while its launch price lasts and for very long-context jobs where Google's GraphWalks numbers are strong.

- Anthropic's own guidance is to start with Opus 5.5 for most workloads and move to Fable 5.2 for demanding reasoning and long-horizon agent work, or when Opus 5.5 at higher effort still falls short.
- Google says Argon remains the right choice for most software engineering and knowledge work, and positions Carbon for the hardest multi-step coding, debugging and research.
- From December 1, Carbon and Opus 5.5 have identical token prices, so the choice comes down to how many tokens each spends on your tasks and which one finishes them. Measure that on your own repository before switching.
- On Google Cloud, Argon is in the table of models available for at least 12 months; Carbon and Hydrogen are listed under shorter availability periods, with no retirement date announced yet.

Outside the API, Google is also rolling Carbon out to Google AI Ultra subscribers in Google Antigravity, the Gemini app and AI Mode in Search. Our related posts below cover Carbon's launch pricing, Fable 5.2's price cut and the lighter tier, Gemini 4 Hydrogen against Claude Sonnet 5.5.

## FAQ

How much will Gemini 4 Carbon cost after November 30? $4 per million input tokens and $20 per million output tokens from December 1, 2026, the same as Claude Opus 5.5\. Through November 30 it is $2 and $10.

How much does Claude Fable 5.2 cost? $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1\. Cache hits are $0.20 per million, and fast mode is $16 and $80.

Which scores highest on the Artificial Analysis Intelligence Index? Fable 5.2 at 65 (1st of 92), then Carbon at 60 (7th) and Opus 5.5 at 59 (8th), as of November 18.

## Related on claudefox

- [Gemini 4 Carbon is available to paying API developers, with launch pricing that doubles after November 30](https://claudefox.com/gemini-4-carbon-available-api-pricing/)
- [Fable 5.2 pricing drops to $8/$40 per million tokens](https://claudefox.com/fable-5-2-pricing-drops-to-8-40-per-million-tokens/)
- [Anthropic launches Claude Fable 5.2 and Mythos 5.2](https://claudefox.com/anthropic-launches-claude-fable-5-2-and-mythos-5-2/)
- [Gemini 4 Hydrogen vs Claude Sonnet 5.5 for coding: price, speed and benchmarks](https://claudefox.com/gemini-4-hydrogen-vs-claude-sonnet-5-5/)

## What we have not tested

We have not run any of these models on our own tasks, so nothing here is a hands-on result. The benchmark rows were chosen and published by Google, Anthropic or Artificial Analysis. If we add our own measurements, we will date them on this page.

Source: [Gemini 4 Carbon is now available (Google, Nov 10, 2026)](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/?ref=claudefox.com), [Gemini models: performance tables (Google DeepMind)](https://deepmind.google/models/gemini/?ref=claudefox.com), [Gemini API pricing (Google)](https://ai.google.dev/gemini-api/docs/pricing?ref=claudefox.com), [Gemini thinking levels (Google AI Studio docs)](https://aistudio.google.com/docs/thinking?ref=claudefox.com), [Model versions and lifecycle (Google Cloud)](https://docs.cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versions?ref=claudefox.com), [Claude Opus 5.5 (Anthropic)](https://www.anthropic.com/claude/opus?ref=claudefox.com), [Claude Fable 5.2 (Anthropic)](https://www.anthropic.com/claude/fable?ref=claudefox.com), [Claude models overview (Anthropic docs)](https://platform.claude.com/docs/en/about-claude/models/overview?ref=claudefox.com), [Claude API pricing (Anthropic docs)](https://platform.claude.com/docs/en/about-claude/pricing?ref=claudefox.com), [Gemini 4 Carbon (Artificial Analysis)](https://artificialanalysis.ai/models/gemini-4-carbon?ref=claudefox.com), [Claude Opus 5.5 (Artificial Analysis)](https://artificialanalysis.ai/models/claude-opus-5-5?ref=claudefox.com), [Claude Fable 5.2 (Artificial Analysis)](https://artificialanalysis.ai/models/claude-fable-5-2?ref=claudefox.com), [Gemini 4 Argon (Artificial Analysis)](https://artificialanalysis.ai/models/gemini-4-argon?ref=claudefox.com)

This story first appeared in the [claudefox daily briefing for November 18, 2026](https://claudefox.com/daily-2026-11-18/).