TL;DR
Anthropic released Claude Fable 5.2 on November 9, and Google opened Gemini 4 Carbon to paying users the next day, followed by Gemini 4 Hydrogen on November 12. Artificial Analysis’s published results show Fable ahead on its overall index and some coding and document tests, while Gemini models lead on several other tasks and cost less per task. The results do not establish a best choice for every team.
Anthropic released Claude Fable 5.2 on November 9, followed by Google’s paid Gemini 4 Carbon on November 10 and Gemini 4 Hydrogen for all users on November 12. In results published by Artificial Analysis, Fable 5.2 leads the overall index at its highest effort setting, while Gemini models lead several individual workplace tests and cost less per benchmark task.
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
At the High effort setting, Artificial Analysis gives Fable 5.2 an Intelligence Index score of 62, compared with 60 for Gemini 4 Carbon. The reported cost per index task is $9.08 for Fable and $5.48 for Carbon. At Fable’s Max setting, its score rises to 65 and the task cost to $26.54. The comparison was accessed on November 17; the figures describe the benchmark’s measured tasks, not a guaranteed bill for a company’s own workload.
Results vary by test. Fable 5.2 scores higher on coding in open-source repositories (49.1% versus 40.6%) and long-context reasoning (83.3% versus 79.3%). Carbon leads on terminal tasks (65.2% versus 59.6%), professional documents (28.4% versus 21.8%) and business software automation (77.8% versus 71.4%). These are Artificial Analysis benchmark results, not evidence that either model will perform identically in every team’s tools or workflow.
The source also reports a reliability trade-off in the same independent runs. Fable answered more questions correctly, with 77.2% accuracy against Carbon’s 58.3%. But its reported hallucination rate was 58.0%, compared with 11.4% for Carbon. Those figures concern the cited test and should not be read as a universal rate for all requests. The source says lower is better for cost and hallucination rate.

Choosing Models by Workload and Cost
For teams, the comparison puts task performance, answer reliability and operating cost side by side. A group that spends much of its time on repository coding or long documents may find Fable’s benchmark lead relevant. Teams focused on terminal work, document production or business software automation may want to examine Carbon’s results. The reported scores offer a starting point for evaluation, rather than a substitute for testing with the team’s own work.
Cost differences may affect which model a team tries first or uses at scale. Artificial Analysis reports a higher cost per index task for Fable at High than for Carbon. Google’s listed token prices for Carbon and Argon are also scheduled to increase on December 1, from $2 per million input tokens and $10 per million output tokens to $4 and $20, respectively. Hydrogen is listed at $1.50 per million input tokens and $7.50 per million output tokens. Token rates alone do not determine total cost: the source notes that the tokens used for a task’s reasoning also matter.
The reliability figures have practical relevance where staff depend on model responses to inform decisions. Fable’s higher accuracy and higher reported hallucination rate describe different measures, not a simple ranking. Teams need to decide how their work handles uncertainty, and whether people review outputs before acting on them. The benchmark does not establish the level of review any particular organization needs.
Products Worth Considering

Sony WH-1000XM6 Wireless Noise-Cancelling Headphones (Black)

Samsung Galaxy Tab A11+ Kids Edition, 128GB, Royal Blue

Amazon Kindle Scribe (16GB) with Premium Pen, Tungsten
As an affiliate, we earn on qualifying purchases.
Three Releases in Four Days
The launch sequence began with Claude Fable 5.2 on November 9. Google opened Gemini 4 Carbon to paying users on November 10, then released Gemini 4 Hydrogen to all users, including through Google Search, on November 12. Gemini 4 is a family: the supplied source describes Carbon as Google’s most capable reasoning model and Argon as its efficient frontier option, which Google recommends for most work. Hydrogen is described as designed for rapid execution across varied agent workflows.
The source reports a one-million-token context window for Fable 5.2 and all three Gemini 4 models. It says Fable supports text and images, as do Gemini 4 Argon; Carbon and Hydrogen are reported to accept text, images, speech and video. These are capabilities attributed to the source’s cited model information, not findings from the head-to-head benchmark figures listed above.
Artificial Analysis’s comparison draws on several named tests, including FrontierCode-AA 2.0, Terminal-Bench 4.0, GDP.pdf, AutomationBench-AA and AA-LCR v1.1. The results are dated to the model pages accessed November 17, 2026. Anthropic’s advice, as quoted in the source, is to start with Claude Opus 5.5 for most work; Google calls Argon “the ideal choice for the majority of software engineering and knowledge work.” Those are vendor recommendations, distinct from the independent test results.
““the ideal choice for the majority of software engineering and knowledge work.””
— Google, describing Gemini 4 Argon
Limits of the Published Comparison
The supplied material does not specify how closely the benchmark tasks match any particular company’s workload, or how results change with different prompts, software integrations and review practices. It also does not provide a complete account of test methodology or sample sizes. The published figures therefore support a task-by-task comparison, but do not settle which model a team should deploy.
The hallucination and accuracy results are reported for specific Artificial Analysis runs. The source does not establish how those rates apply across other kinds of questions or real business use. Nor does it report what proportion of a team’s work each benchmark represents, so the scores cannot by themselves predict a team’s overall productivity or savings.
Pricing also depends on use. The source gives token rates and benchmark task costs, but not a forecast of a particular organization’s monthly bill. Carbon and Argon’s introductory prices are set to expire on December 1; the material does not state whether Google will extend them or change other pricing terms. The source also does not provide comparable task costs for every Gemini 4 variant at every setting.
Teams Can Test Their Own Work
The next useful step for teams weighing the models is a side-by-side trial using representative tasks: the codebases, terminal work, documents and business software their staff actually use. A trial can track completion quality, correction time, response reliability and token use. Those measures would help teams connect benchmark differences to their own work and costs.
Pricing is another near-term milestone. Google’s listed introductory rates for Gemini 4 Carbon and Argon are due to rise on December 1, 2026, according to the source. Organizations considering either model can compare their expected token use against both sets of rates. The supplied material does not identify further scheduled releases or changes to the benchmark comparison.
For now, the published evidence points to different advantages rather than a single overall choice for all teams: Fable leads the index at higher effort, while Carbon and other Gemini variants lead some task tests and have lower reported costs per index task. How those findings translate to a team remains an open question until it evaluates the models on its own work.
Key Questions
Which model scores highest on the Artificial Analysis Intelligence Index?
Claude Fable 5.2 scores 65 at Max effort, the highest figure reported in the supplied comparison. At High effort it scores 62, compared with 60 for Gemini 4 Carbon.
Which model costs less per index task?
At the compared settings, Gemini 4 Carbon costs $5.48 per task and Fable 5.2 at High costs $9.08, according to Artificial Analysis. These are benchmark task costs, not estimates for every company workload.
Does Fable 5.2 lead on every task?
No. In the supplied results, Fable leads on repository coding and long-context reasoning. Carbon leads on terminal tasks, professional documents and business software automation.
When do Gemini 4 Carbon and Argon prices change?
The source says Google’s introductory token rates apply through November 30, 2026. From December 1, listed rates rise from $2 to $4 per million input tokens and from $10 to $20 per million output tokens.
Source: Thorsten Meyer AI
