In this article
On November 9, 2026 Anthropic released Claude Fable 5.2 together with its sibling Claude Mythos 5.2. In short: the price fell 20% to $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1; output is about 30% faster; the model beats its predecessor on every benchmark the company published; and it blocks about 55% fewer requests. Even so, Anthropic itself recommends starting with Opus 5.5 for everyday coding and knowledge work, and keeps Fable 5.2 for large, long-running jobs. So the decision for your company is not «always the newest», but which tasks justify the more expensive model.
What is new in Fable 5.2, in numbers
These are Anthropic's published results from the model page, against Fable 5.1 and Opus 5.5:
- Agentic coding (FrontierSWE v2): 69.9% against 56.3% for Fable 5.1, 62.3% for Opus 5.5 and 65.5% for GPT-6 Astra.
- Agentic science (Terminal-Bench-Science 0.1): 74.6% against 52.6% for its predecessor.
- Agentic web search (WANDR): 82.1% against 68.7%.
- Computer use (OSWorld 2.1): 89.3% against 83.5%.
- Real remote work (Remote Labor Index): 41.3% against 17.9%, more than double.
- Reading charts without tools (Chartography): 73.4% against 44.8%.
- Humanity's Last Exam with tools: 74.2% against 65.6%.
To be fair, the same table shows the model does not lead everywhere: Gemini 4 Carbon beats it on business workflow automation (AutomationBench) at 53.9% against 51.4%, and Opus 5.5 edges it on the knowledge-work Elo rating (1800 against 1785). All of these are the company's own numbers; independent measurements have not been published yet.
How it compares with Fable 5.1 and ChatGPT
OpenAI's strongest ChatGPT model today is GPT-6 Astra, which its developer page describes as its most capable model for the most demanding work. Here are the three side by side:
| Item | Fable 5.2 | Fable 5.1 | GPT-6 Astra (ChatGPT) |
|---|---|---|---|
| Input price per million tokens | $8 | $10 | $10 |
| Output price per million tokens | $40 | $50 | $50 |
| Cache read | $0.20 | $0.25 | $1 |
| Agentic coding (FrontierSWE v2) | 69.9% | 56.3% | 65.5% |
| Agentic science (Terminal-Bench-Science) | 74.6% | 52.6% | 68.1% |
| Computer use (OSWorld 2.1) | 89.3% | 83.5% | 79.5% |
| Real remote work (Remote Labor Index) | 41.3% | 17.9% | 20.8% |
| Humanity's Last Exam with tools | 74.2% | 65.6% | 57.2% |
| Workflow automation (AutomationBench) | 51.4% | 31.4% | 41.4% |
| Reading charts (Chartography) | 73.4% | 44.8% | 71.0% |
The takeaway: Fable 5.2 is 20% cheaper than GPT-6 Astra on list price, and the gap is wider on agentic work because cache reads cost 5 times less. With GPT-6 Astra, the whole request moves to 2 times the input rate and 1.5 times the output rate once input exceeds 272 thousand tokens, and its context window is about 1.05 million tokens. Note that the benchmark results here are published by Anthropic, not by OpenAI or an independent party.
Price and speed: what changes on the bill
- Standard price: $8 (SAR 30) per million input tokens and $40 (SAR 150) per million output tokens, instead of $10 and $50. Cache reads cost $0.20 per million.
- Fast mode: up to 2.5 times faster generation at double the price: $16 input and $80 output. Available on the Claude Platform, Claude Code, Google Cloud and Microsoft Foundry.
- Fewer tokens: Anthropic says Fable 5.2 at medium effort uses fewer tokens than 5.1 on hard tasks, so the bill drops twice: a lower price and a smaller quantity.
- US-only inference costs 1.1 times the standard price for those who need it.
Agents that work for days, not minutes
The biggest focus of this release is long-running work: the model runs ambiguous engineering work for days, coordinates hundreds and even more than a thousand subagents, tests its own work as it goes, and flags results that do not reproduce. Published customer examples include:
- At Uber, a 10 to 12 hour security audit in which the model sent out about 170 reviewer agents and came back with a write-up, fix pull requests and a policy proposal.
- Discovered Materials made its Quantum ESPRESSO code about 36 times faster on average, taking its 13 benchmark cases from 3.3 hours to under 4 minutes.
These are the customers' own claims, but they show the kind of work the model is built for: a whole project handed over, not a question and an answer.
Safeguards: fewer blocks, with conditions to know before you integrate
- Fable 5.2 blocks about 55% fewer requests than Fable 5.1, one of the most common user complaints about earlier releases.
- Cybersecurity, biology and chemistry requests that trigger the safeguards are automatically routed to less capable Opus models. API customers must configure the new Fallback API so their service does not stall.
- Using Fable models requires Anthropic to retain data for 30 days by default for safety monitoring, which matters to any company that passes customer data through it.
- Mythos 5.2 is the same model without these limits, available only to verified organizations through the cyber and life sciences verification programs.
Fable 5.2 or Opus 5.5?
Anthropic's official page says it plainly: for most everyday coding and knowledge work, start with Opus 5.5. Fable 5.2 is available on the Pro, Max, Ultra, Team and Enterprise plans, on the API as claude-fable-5-2, and on Amazon, Google Cloud and Microsoft Foundry. We covered the technical differences between the two families in our article Opus 5.5 vs Fable 5.1, and most of them still hold with 5.2.
- Choose Opus 5.5 for chat, content writing, short tasks and everyday automation.
- Choose Fable 5.2 for a large multi-repository software project, a full review, or analysis that needs hours of independent work.
The Origami view
We use Claude models every day to build our clients' systems, and the lesson we have learned is that the difference on the bill comes from routing tasks to the right model, not from picking one model for everything. A 20% cut in Fable's price makes it easier to justify on heavy jobs, but it does not make it the right choice for every customer reply.
For Saudi businesses, we think the most important parts of this release are two items that do not make the headlines: 30-day data retention, and automatic routing to a weaker model when a safeguard fires. Both need a clear decision before the model is wired into a system that holds customer data.
The bottom line: what to do now
- If you use Fable 5.1 through the API, move to
claude-fable-5-2: cheaper and stronger on the same terms. - Review any short tasks you run on Fable and move them to Opus 5.5.
- Configure the Fallback API before going live so your service does not stop when a request is blocked.
- If your company passes sensitive data, read the 30-day retention condition first, or contact us to review the integration with you.
Sources
Frequently asked questions
What is new in Claude Fable 5.2 compared with Fable 5.1?
A 20% lower price ($8 and $40 per million tokens instead of $10 and $50), output about 30% faster, higher scores on every benchmark Anthropic published, such as 69.9% on FrontierSWE v2 against 56.3%, and about 55% fewer blocked requests.
How much does Claude Fable 5.2 cost on the API?
$8 per million input tokens and $40 per million output tokens. Fast mode costs $16 and $80 and is up to 2.5 times faster.
Is Fable 5.2 better than Opus 5.5 for my company?
For large, long-running tasks, yes: it leads on most benchmarks. But Anthropic itself recommends starting with Opus 5.5 for everyday coding and knowledge work, because it is cheaper and enough for most tasks.
Is Claude Fable 5.2 better than ChatGPT?
By Anthropic's numbers, Fable 5.2 beats GPT-6 Astra, the strongest ChatGPT model, on almost every published benchmark, for example 69.9% against 65.5% in agentic coding, and it is 20% cheaper on list price. These are the company's own results, so it is best to test both models on your own tasks.







