What Is Claude Fable 5 | A Deep Dive into Anthropic's Latest Frontier Model: Performance, Pricing, and How to Use It
A comprehensive look at Claude Fable 5, Anthropic's latest model released in June 2026. Covers benchmark performance in coding, knowledge work, and science, how it compares to Mythos 5, API pricing, and usage guidance — all backed by official data.
On June 9, 2026, Anthropic released its new frontier model Claude Fable 5. At the same time, a sister model with some safety guardrails removed — Claude Mythos 5 — became available on a limited basis. Both share the same underlying model: Fable 5 is the general-availability version with safety mechanisms in place, while Mythos 5 is the version with guardrails removed for authorized users.
This article covers Claude Fable 5's benchmark performance, capabilities in coding, knowledge work, and scientific research, API pricing, and how to get started — all based on Anthropic's official announcement data. The model ID is claude-fable-5.
All data and figures in this article are sourced from the official Anthropic announcement (June 9, 2026).
[IMPORTANT] Fable 5 / Mythos 5 Access Was Suspended on June 12, 2026
Access was originally planned to be free for subscribers through June 22, 2026, but on June 12, 2026 (5:21 PM ET), the U.S. government issued an export control directive, immediately suspending access to Fable 5 and Mythos 5 ahead of that scheduled date.
- Who is affected: All users of Fable 5 / Mythos 5, including users outside the U.S. and Anthropic employees based outside the United States.
- Reason: The government cited national security concerns around jailbreak techniques for Fable 5. Anthropic disputes the measure, stating that this is a minor, known vulnerability also findable in competing models.
- Anthropic's response: While the company is complying with the government's legal directive, it argues that "finding a narrow jailbreak is not a reason to recall a widely deployed model — applying this standard would effectively halt new model deployments across the entire industry."
- Other models: Anthropic models other than Fable 5 / Mythos 5 (such as Opus 4.8) are unaffected and remain available.
Accordingly, the pricing and availability details covered later in this article (free access through June 22, credit-based access from June 23 onward, etc.) reflect the original announcement and are no longer accessible due to the suspension. Please check the official Anthropic announcement for the latest status.
Source: Anthropic, "An update on Fable 5 and Mythos 5 access" (June 12, 2026)
What Is Claude Fable 5?
Claude Fable 5 is Anthropic's latest frontier model, which the company describes as "the most capable model we've ever made available to the public." It achieves state-of-the-art (SOTA) performance on nearly every benchmark tested and surpasses its predecessor, Claude Opus 4.8, across a wide range of domains including coding, knowledge work, vision, and scientific research.
The relationship between Fable 5 and Mythos 5 is as follows:
- Claude Fable 5: The general-availability version with safety guardrails. Available to everyone via the Claude API and Claude apps.
- Claude Mythos 5: A version with cybersecurity and bio/chemical safety guardrails removed from the same base model. Available only to authorized users such as Project Glasswing partners and select biomedical researchers.
Because greater capability also raises the risk of misuse, Anthropic has explicitly stated that "without its safety mechanisms, Fable 5 would have capabilities that could be exploited in areas such as cybersecurity," which is why the public version includes those guardrails.
Claude Fable 5 Benchmark Performance
The table below shows the official benchmark comparison between Claude Fable 5 / Mythos 5 and other leading models.
Here are the highlights compared to Claude Opus 4.8 (scores reflect the higher of Mythos 5 and Fable 5):
| Benchmark | Domain | Fable/Mythos 5 | Opus 4.8 |
|---|---|---|---|
| SWE-Bench Pro | Agentic coding | 80.3% | 69.2% |
| Terminal-Bench 2.1 | Agentic coding | 88.0% | 82.7% |
| FrontierCode (Diamond) | Coding | 29.3% | 13.4% |
| GDPval-AA | Knowledge work | 1932 | 1890 |
| GDP.pdf | Knowledge work (vision) | 29.8% | 22.5% |
| Humanity's Last Exam (with tools) | Multi-domain reasoning | 64.5% | 57.9% |
| ExploitBench | Cybersecurity | 78.0% | 40.0% |
| HealthBench Professional | Healthcare | 66.0% | 56.9% |
Across comparisons that include GPT-5.5 and Gemini 3.1 Pro, Fable 5 records the highest scores on most categories. Additionally, it becomes the first frontier model to exceed 90% on a core analytics benchmark (+10 points over Opus), and completes spreadsheet tasks 25–30% faster than Opus 4.8, marking a significant improvement in real-world knowledge work utility.
Claude Fable 5 Coding and Agentic Performance
Claude Fable 5 is particularly strong at long-horizon autonomous coding tasks (agentic coding).
- Large-scale migrations in a single day: At Stripe, it completed the migration of a 50-million-line Ruby codebase in one day — a task that normally takes two months.
- Top scores on FrontierCode / FrontierBench: It recorded the highest scores on Cognition's frontier coding evaluations.
- It compresses months of engineering work into days, dramatically boosting development productivity.
Here are summary quotes from evaluations by leading developer tool companies:
- Cursor (CEO Michael Truell): State-of-the-art model on CursorBench.
- GitHub (CPO Mario Rodriguez): Surpasses traditional benchmarks on long-horizon tasks — a major step forward for developers.
- JetBrains (Matt Colyer): The strongest Claude model we've tested. A clear leap forward in agentic coding.
- Vercel (CTO Fabian Hedin): Generated in one shot an app that previously required 100 prompts.
For practical coding techniques using Claude, see the Complete Guide to Using Claude Code.
Knowledge Work, Vision, and Long Context
- Vision: Achieves state-of-the-art performance on visual tasks such as extracting precise values from scientific figures and reconstructing web app source code from screenshots. It also made headlines by completing the game "Pokémon FireRed" using a vision-only approach.
- Long context and memory: Can maintain focus across millions of tokens. In the game "Slay the Spire" using file-based persistent memory, it delivered 3× the performance of Opus 4.8 and reached the final stage 3× more often.
- Finance and specialist domains: Scored highest on Hebbia's financial benchmark, rated at senior-level reasoning.
Scientific Applications (Mythos 5)
Mythos 5, with some safety guardrails removed, is being used by authorized biomedical researchers in drug discovery and life science applications.
- Drug discovery: Accelerated certain steps of protein design approximately 10-fold. Generated promising drug candidates for 9 out of 14 target proteins.
- Novel hypothesis generation: The first model to consistently generate novel and compelling scientific hypotheses. In blinded comparisons, it showed approximately 80% preference over Opus-class models, and a novel hypothesis about E. coli was independently confirmed by separate research.
- Genomics: Integrated millions of single-cell data points from 138 animal species to build a proprietary machine learning model at 1/100th the scale of a Science-published method, while surpassing its results.
Figure: Protein complexes designed by Mythos 5, targeting immune checkpoints, growth factor/receptor signaling, neurodegeneration, muscle disease, and more challenging structural targets.
Safety and Alignment
Given its expanded capabilities, Anthropic has invested heavily in safety evaluation.
- Alignment: Misaligned behavior in automated alignment evaluations is at a low level, on par with Claude Opus 4.8.
- Cybersecurity guardrails: Against 30 publicly available jailbreak techniques, zero harmful single-turn requests were complied with. Safety guardrails are triggered in fewer than 5% of sessions on average, and more than 95% of Fable sessions complete without falling back to Opus 4.8.
- Red-teaming: Over 1,000 hours of external bug bounty testing found no universal jailbreak. External organizations also failed to find a universal jailbreak during long-horizon agentic tasks (the UK AISI made some progress within the initial testing window), and Fable 5 demonstrated greater jailbreak resistance than any previous generally available model.
Mythos 5, with its guardrails removed, possesses capabilities that could be misused — and is therefore provided only to a limited set of trusted users, such as Project Glasswing partners and select biomedical researchers.
Claude Fable 5 Pricing and Availability
⚠️ The following reflects the state of Anthropic's announcement as of June 9, 2026. As noted above, access to Fable 5 / Mythos 5 was suspended by U.S. government directive on June 12, 2026, and the availability details below are no longer active.
Claude Fable 5 API pricing is as follows:
| Item | Price (per million tokens) |
|---|---|
| Input tokens | $10 |
| Output tokens | $50 |
Availability is as follows:
- Claude API: Fully available from day one (model ID
claude-fable-5). - Subscriptions: Available at no extra charge for Pro / Max / Team / seat-based Enterprise plans from June 9 through June 22, 2026.
- From June 23 onward: Usage credits required. Standard availability expected to return "as soon as possible."
- Pay-as-you-go Enterprise: Fully available from day one.
How to Use Claude Fable 5
Claude Fable 5 is available through the following channels:
- Claude API: Submit requests specifying the model ID
claude-fable-5. - Claude apps (Web / Desktop / Mobile): Select the model under eligible plans.
- Enterprise: Available from day one under pay-as-you-go plans.
ai-best-search also features detailed information on Claude and Claude Code.
Frequently Asked Questions About Claude Fable 5
Is Claude Fable 5 available in English?
Yes. Claude Fable 5 supports multiple languages including English, available via the Claude apps and Claude API. It handles tasks such as writing, summarization, translation, and coding assistance in English.
What is the difference between Claude Fable 5 and Claude Opus 4.8?
Fable 5 is the next-generation model, succeeding Claude Opus 4.8. It outperforms Opus 4.8 on key benchmarks such as SWE-Bench Pro (80.3% vs. 69.2%) and Terminal-Bench 2.1 (88.0% vs. 82.7%), with improved performance on long-horizon agentic tasks and scientific applications.
Can I use Claude Fable 5 for free?
During June 9–22, 2026, it was available at no extra charge for Pro / Max / Team / seat-based Enterprise plan subscribers. From June 23 onward, usage credits are required. The API is pay-as-you-go at $10 per million input tokens and $50 per million output tokens.
What is the difference between Claude Fable 5 and Claude Mythos 5?
Both are built on the same base model. Fable 5 is the general-availability version with safety guardrails intact, while Mythos 5 is a version with cybersecurity and bio/chemical safety guardrails removed, provided only to authorized users.
What is the model ID for Claude Fable 5?
The model ID is claude-fable-5. Use this ID when making requests via the Claude API.
Summary
- Claude Fable 5 is Anthropic's most capable publicly available frontier model (released June 9, 2026).
- It achieves state-of-the-art performance across coding, knowledge work, and science — including SWE-Bench Pro at 80.3% and Terminal-Bench 2.1 at 88.0%.
- Sister model Mythos 5 is a limited-access version with guardrails removed, delivering results in drug discovery and scientific research.
- API pricing: $10 input / $50 output per million tokens. Model ID:
claude-fable-5.
Get a clear picture of what Claude Fable 5 can do and start applying it to your work.