Learn  /  What's new

Claude Fable 5.2 vs Gemini 4 Carbon: the new top AI models compared on scores, price and speed

In one week the two strongest AI models both became available to the public. One is clearly smarter, the other is much cheaper per task, and for most data work neither is the obvious choice. Here is how they compare.

What's new8 min readUpdated 16 November 2026
AIWhat's new
All guides and cheat sheets

Two releases changed the top of the AI leaderboard in the week of 9 November 2026. Anthropic launched Claude Fable 5.2 on 9 November, and Google made Gemini 4 Carbon generally available on 10 November. Both are the "big brother" models of their families: slower and more expensive than the everyday models, built for the hardest problems. Our November model rankings cover the rest of the field.

65Index scoreClaude Fable 5.2 at max effort
60Index scoreGemini 4 Carbon at high effort
3.4Times the cost per taskFable 5.2 vs Carbon
20Percent price cutFable 5.2 vs Fable 5.1

The scores

Artificial Analysis runs every model on the same ten tests, most of them private so that labs cannot train on the answers. On version 5.0.1 of its index, Fable 5.2 at maximum effort leads clearly. Even on its "high" setting it scores 62, still above every other model.

Artificial Analysis Intelligence Index v5 (higher is better)
Claude Fable 5.2 (max)65
Claude Fable 5.2 (high)62
Gemini 4 Carbon (high)60
Claude Opus 5.5 (max)59
GPT-6 Astra (max)57
Claude Fable 5.2 (low)54
Claude Fable 5.1 (max)49

Source: Artificial Analysis via OfficeChai, 10 November 2026. Effort setting in brackets. Gemini 4 Carbon and Claude Opus 5.5 each gained one point in the v5.0.1 revision.

The total hides where each model wins. Fable 5.2 leads on knowledge and reasoning, and makes up fewer facts. Carbon still wins on agent-style tasks that click through tools and terminals.

TestClaude Fable 5.2Best other model
HLE-Diamond (hard exam questions)80%69%, Gemini 4 Carbon
AA-Omniscience (knowledge without made-up facts)6754, Gemini 4 Carbon
Terminal-Bench-Science73%65%, Gemini 4 Carbon
CritPt v2 (hard reasoning)77%72%, GPT-6 Astra
FrontierCode 2.0 (real-world coding)49%46%, Claude Opus 5.5
AutomationBench (business automation)76%78%, Gemini 4 Carbon
Terminal-Bench 4.060%65%, Gemini 4 Carbon; 67%, Kimi K3.1 Pro

Price and cost per task

Token prices are only half the story. What matters is the bill for finishing a job, and that depends on how many tokens a model spends thinking.

Claude Fable 5.2Gemini 4 Carbon
Input / output per million tokens$8 / $40$2 / $10 until 30 November, then $4 / $20
Cost per index task (max effort)$26.50$7.76
Cached input$0.20 per million tokensNot announced
Context window1 million tokens, text and imagesNot stated by Google
Who can use itAll paid Claude plans and the APIPaid Gemini API, Gemini for business, AI Ultra in the Gemini app

What is new in each model

Claude Fable 5.2

Smartest
  • Highest score on the index, and the best at avoiding made-up facts
  • Up to 30% faster than Fable 5.1, at a 20% lower price
  • Can coordinate a team of agents on one task
  • About three times the cost per task of Carbon
  • Safety filters can send some cyber and biology requests to a different model

Anthropic still calls Claude Opus 5.5 its everyday model; Fable 5.2 is for the hardest problems.

Gemini 4 Carbon

Best value at the top
  • Second on the index at about a third of the cost per task
  • Google reports first place on CursorBench 4.0 and Zapier's AutomationBench
  • /swarm in Google Antigravity (preview) turns one prompt into a team of agents
  • Introductory price ends on 30 November
  • Spends more tokens than Gemini 4 Argon, which Google still recommends for most work

The strongest choice for agentic work, especially before the price rises.

Google also shipped a fast sibling on 12 November: Gemini 4 Hydrogen, built for everyday coding and agent steps. Google says it is about 2.5 times faster than Gemini 3.8 Flash at roughly the same price, and it can call several tools at once.

Where data teams can use them today

Availability in data platforms

  1. Databricks: Claude Fable 5.2

    Available through Unity Gateway and the Foundation Model APIs.

  2. Databricks and BigQuery: Gemini 4 Carbon

    On Unity Gateway, and generally available in BigQuery's generative AI functions.

  3. Databricks and BigQuery: Gemini 4 Hydrogen

    Same two places, two days later.

  4. SnowflakeNow

    Its latest model addition, on 6 November, brought Claude 5.5 and GPT-6 models to Cortex; check the Cortex model list for the new ones.

Inside SQL, you rarely need the biggest model. Classifying tickets or pulling fields from documents works well with small, cheap models; see AI functions in SQL compared. For yes/no and multiple-choice questions at scale, Google's new Gemini Decisions API is cheaper still.

Which one to use

  1. 1

    Everyday analysis, SQL and Python

    Stay with Claude Opus 5.5, Claude Sonnet 5.5 or Gemini 4 Argon. Both makers say so themselves.

  2. 2

    The hardest questions, where being right matters more than cost

    Claude Fable 5.2, on high effort first and max only when needed.

  3. 3

    Long agent runs and automation

    Gemini 4 Carbon, ideally before 1 December while the introductory price lasts.

  4. 4

    Fast agent steps and high volume

    Gemini 4 Hydrogen, or a small model inside your warehouse.

  5. 5

    Before switching

    Re-run five real tasks from your own work on both models, and compare the answers and the bill.

Check your understanding

1/3Which model leads the Artificial Analysis Intelligence Index after 9 November?

2/3What happens to Gemini 4 Carbon's price on 1 December?

3/3Why can a cheaper token price still give a bigger bill?

More: the best AI models in November 2026, Gemini 4 Carbon explained and ChatGPT, Claude or Gemini for data analysis.

Sources: SD Times on Claude Fable 5.2 and Mythos 5.2 (9 November 2026); OfficeChai's report of the Artificial Analysis results (10 November 2026); Google, "Gemini 4 Carbon is now available" (10 November 2026); ThursdAI's November 2026 release list for Gemini 4 Hydrogen; Databricks and BigQuery release notes (9 to 12 November 2026); Snowflake feature updates (6 November 2026). Benchmark and speed claims come from the companies and Artificial Analysis.

Alessandro Ecclesie AgazziWritten by Alessandro Ecclesie Agazzi, analytics engineer in London. Updated 16 November 2026. About me

Found this useful?

New guides and "what's new" explainers on data and AI land here regularly. Follow along on LinkedIn, or share this one with a colleague.