🔥 Trending on HN

Claude Fable 5.2 leads one benchmark, but speed and price tell another story

2 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
Artificial Analysis Intelligence Index(Artificial Analysis Intelligence Index)

A combined score from nine different AI evaluations.

Hacker News(Hacker News)

An online community focused on technology news.

context window(context window)

The amount of text an AI can handle together.

What happened

Claude Fable 5.2, an AI model from Anthropic, reached 65 on the Artificial Analysis model page. The page ranks it first among 74 models in its comparison set. It was released in November 2026. It accepts text and images, then produces text.

The story received 357 points and 259 comments on Hacker News. Those numbers show attention in that community. They do not prove that the score is correct.

What the 65 measures

The score belongs to Artificial Analysis Intelligence Index version 5.0.1. Its methodology page says the index combines nine evaluations. They cover knowledge work, coding, science, and general knowledge. Each category counts for 25% of the total.

The evaluations include simulated business workflows, long professional documents, software repositories, terminal tasks, science problems, factual questions, and hallucination checks. The suite is mainly text-based and English-language. Some tasks use images. Artificial Analysis measures multilingual and speech abilities separately.

This scope matters. A broad score can show performance across several kinds of work. It can also hide weak spots inside the total. A model might excel at coding while struggling with another category. The overall number cannot show every detail.

Why the result matters

A high score can make a model interesting for difficult knowledge work. Claude Fable 5.2 is also listed as a reasoning model. The page gives it a one-million-token context window. That describes its reported capacity. It does not guarantee equally strong results on every long task.

The ranking is useful only inside its stated comparison. It is not a claim that Fable wins every AI task. It is also not a replacement for testing a model with a person’s own documents and workflow.

The trade-offs

The same page reports 44.6 minutes per Intelligence Index task. The median time for comparable models is 12.7 minutes. The first answer token appears after 221.21 seconds. Output speed is 102 tokens per second.

Fable’s listed price is $8 per million input tokens and $40 per million output tokens. The comparison medians are $2 and $10. Artificial Analysis estimates an average cost of $26.54 per index task. Fable generated 275 million output tokens during the index. The median was 72.8 million. The headline score is high, but the page also shows higher cost, slower work, and longer answers.

What is confirmed, and what is not

The Artificial Analysis page provides the score, rank, test design, speed, and price figures. Hacker News confirms that the topic drew strong attention there. Neither fact proves real-world superiority.

This source set does not show how Fable performs in a particular job. It also does not establish whether another evaluator would get the same ranking. The available data cannot answer whether the extra cost is worth paying.

What to watch next

Readers should look for results from each of the nine evaluations. Real tasks can reveal whether the model’s long answers help or slow people down. Independent checks can test whether the ranking holds elsewhere. The useful question is not only whether Claude Fable 5.2 leads. It is when its extra ability justifies its time and price.

💬 Fable 5.2: a record score, with cost and real-world caveats

The article headline reports a score of 65 on Artificial Analysis. HN commenters see a potentially stronger model, but disagree about whether its extra reasoning is worth the cost, whether the benchmark is decisive, and how well it behaves in real work.

  • On cost, one commenter reported that at maximum effort, output tokens rose from 118k for Fable 5.1 (46k answer and 71k reasoning) to 274k for 5.2 (81k answer and 192k reasoning), with roughly double the cost per task.
  • Another commenter reported these Artificial Analysis table values for 5.2 versus 5.1: xhigh, 62/$14.83 versus 50/$10.85; high, 62/$9.08 versus 47/$7.54; medium, 56/$3.80 versus 43/$5.56; and low, 54/$3.10 versus 39/$4.37. On that reading, 5.2 is both cheaper and higher-scoring at low and medium. These are the commenter’s reported chart readings, not independent verification.
  • A commenter warned that the AA charts default to maximum effort. They considered medium or high more attractive for ordinary use, while max may be useful for showing the model’s limits but poor value for money.
  • In a positive user report, 5.2 handled 3D modeling with simpler, more efficient, and prettier models while still meeting the functional requirements that 5.1 met.
  • Negative user reports point the other way: one user reported drift and repeated false claims, preferring Opus 5.5, while another reported that 5.2 invented results and skipped a required paper or RAG search. These are individual trials, not evidence of a general failure rate.
  • The benchmark itself is disputed. One commenter alleged that AA is funded by AI labs and does not publish its full method; another replied that the methodology is public and that private benchmarks need to stay hidden to reduce gaming.
  • The practical conclusion is conditional: Fable 5.2 may be worthwhile for expensive planning, review, or other high-value tasks where a small reliability gain matters. For routine work, lower effort levels or cheaper models may be the better choice.

initial digest at 259 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.

🔥 Trending on HN

Claude Fable 5.2 scores high, but it is not fast or cheap

📰 Full story: Claude Fable 5.2 leads one benchmark, but speed and price tell another story

Claude Fable 5.2 leads one AI ranking. Its high score comes with longer waits and higher costs.

1 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
Intelligence Index(Intelligence Index)

A score that combines several AI tests.

token(token)

A small unit used to count text for an AI.

Hacker News(Hacker News)

A technology news community.

💡 The gist

  • Anthropic made Claude Fable 5.2.
  • Artificial Analysis ranked it first among 74 models.
  • Its score is high, but its speed and price are also high.

Anthropic is the AI company behind Claude. Artificial Analysis is an AI testing site. Its page gives Fable 5.2 an Intelligence Index score of 65. The page lists it first among 74 models. This ranking belongs to that comparison list. It is not a promise that Fable wins every task.

The score combines nine tests. The tests cover knowledge work, coding, science, and general knowledge. Each group counts for one quarter. Some tasks use documents or images. Others use simulated work apps, code repositories, or computer terminals. The index mainly uses English and text. Other tests measure languages and speech.

This design makes the number useful. One test can miss important skills. Several tests can show a wider pattern. A combined score can still hide weak spots. Readers need to ask which tasks went well.

Speed matters too. One index task took Fable 44.6 minutes. The middle result for comparable models was 12.7 minutes. A long answer may help a difficult task. Waiting 44.6 minutes may hurt a live workflow.

Price matters as well. Fable costs $8 per million input tokens. It costs $40 per million output tokens. The comparison medians are $2 and $10. A million tokens means a very large amount of text. It is not one ordinary word.

Hacker News showed strong interest. The post had 357 points and 259 comments. These numbers measure attention in that community. They do not prove the score is correct.

The safest conclusion is limited. Fable 5.2 leads this Artificial Analysis list. It also uses more time and money. Next, readers should watch detailed results, independent checks, and real work examples. Those details can show when the high score is useful.

💬 Fable 5.2: powerful, expensive, and debated

The article says Fable 5.2 reached 65 on Artificial Analysis. Comments praise its potential, but also question its price, reliability, and the meaning of the score.

  • One commenter reported that maximum-effort 5.2 used 118k to 274k output tokens compared with 5.1, and cost about twice as much per task.
  • Another commenter read the AA table as showing that low and medium settings were cheaper and scored higher than the same settings on 5.1. So the setting changes the value a lot.
  • Because AA shows maximum effort by default, some commenters said max looks unattractive for everyday work, while medium or high may be more practical.
  • User experiences were mixed: one person reported better 3D models, while others reported drifting answers, made-up results, or skipped document searches.
  • Commenters also disagreed about whether AA’s published methodology is enough to trust. The score is useful evidence, but not the whole answer.

initial digest at 259 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.

🔥 Trending on HN

Claude Fable 5.2 did very well on tests

📰 Full story: Claude Fable 5.2 leads one benchmark, but speed and price tell another story

Claude Fable 5.2 got a high score on many AI tests.

1 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
Anthropic(Anthropic)

The company that makes Claude.

Artificial Analysis(Artificial Analysis)

A website that tests AI systems.

Hacker News(Hacker News)

A place where people discuss technology news.

Anthropic is a company that makes Claude. Claude Fable 5.2 is an AI system. It reads questions and writes answers.

Artificial Analysis is a website that tests AI systems. It tested Fable with many kinds of questions. Fable scored 65. It ranked first among 74 models on that list.

The score joined nine different tests. The tests asked about knowledge, coding, science, and general questions. A high score means Fable did well on those tests. It does not mean Fable is best at everything.

Fable was not very quick. One test task took 44.6 minutes. Similar models took 12.7 minutes in the middle. More time can be hard when someone needs an answer quickly. Fable also costs more than many models.

Hacker News is a technology news community. The story got 357 points and 259 comments there. That means many people noticed it. It does not mean the score is true.

The news has two parts. Fable got a very high test score. Fable also needs more time and money. People should test it on real work next. They should check whether the extra help is worth waiting for.

💬 Fable 5.2 may be very clever, but it can cost a lot

The article says it got 65 points in a test. People in the comments did not all have the same experience.

  • One user reported that the strongest setting thought much more and cost about twice as much as 5.1.
  • Another user reported that weaker settings were cheaper and still worked better than 5.1. Choosing the setting matters.
  • Some users reported prettier 3D work. Others reported confused answers or missing important searches. Those are personal experiences, not a final verdict.
  • People also disagreed about how much to trust the test. A high score alone cannot tell everyone which model to use.

initial digest at 259 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.

Sources