Anthropic launched Claude Fable 5 as its most capable generally available model: a “Mythos-class” model with safety safeguards applied for public use. If you already build with Claude Opus 4.8, treat Fable 5 as the new premium frontier tier above it. This guide focuses on what developers can actually do with it: when to use it, how it differs from Claude Mythos 5, how pricing works, and how to test the claude-fable-5 API before production.
TL;DR
Claude Fable 5 launched on June 9, 2026 as Anthropic’s new public frontier model. It is a “Mythos-class” model made safe for general use, supports long-running autonomous tasks across millions of tokens, and is available through the Claude API with the model ID:
claude-fable-5
API pricing:
| Token type | Price |
|---|---|
| Input | $10 / 1M tokens |
| Output | $50 / 1M tokens |
Use Fable 5 when your workload involves long-horizon coding, codebase migration, deep document analysis, complex agent loops, or multi-step reasoning where cheaper models tend to lose context.
What is Claude Fable 5?
Claude Fable 5 is Anthropic’s public frontier model from the June 2026 release. Anthropic describes it as “a Mythos-class model made safe for general use.”
That phrase matters:
- Mythos-class refers to the capability tier.
- Fable 5 is the generally available model.
- Mythos 5 is the restricted version with some safeguards lifted.
Fable 5 and Mythos 5 share the same underlying model. The difference is the safety layer applied to Fable 5 so it can be made available through normal API and Enterprise access.
For developers, the important implementation detail is simple: use this model string in Claude API requests:
{
"model": "claude-fable-5"
}
If you already use Anthropic models, upgrading is mostly a model swap plus validation. The migration path is similar to the one described in the guide to using the Claude Opus 4.8 API. Anthropic also lists Fable 5 in its models overview documentation.
Claude Fable 5 vs Claude Mythos 5
Anthropic launched two related models on June 9, 2026:
- Claude Fable 5: public, available through the Claude API.
- Claude Mythos 5: restricted, available only through controlled programs.
Both use the same underlying model. Fable 5 includes full safeguards. Mythos 5 has some safeguards lifted for specific trusted-access use cases.
| Aspect | Claude Fable 5 | Claude Mythos 5 |
|---|---|---|
| Availability | Public via Claude API | Restricted, not public |
| Underlying model | Same as Mythos 5 | Same as Fable 5 |
| Safeguards | Full safeguards on | Some safeguards lifted |
| Who can use it | API and consumption-based Enterprise customers | Project Glasswing cyberdefenders, infrastructure providers, later biology researchers, Mythos Preview upgraders |
| Pricing | $10 input / $50 output per 1M tokens | $10 input / $50 output per 1M tokens |
| API model ID | claude-fable-5 |
Not generally exposed |
The practical takeaway: if you are building a product, internal tool, coding agent, or analysis workflow today, Fable 5 is the model you can actually use. Mythos 5 is not a normal self-serve API product.
What Claude Fable 5 is good for
Fable 5’s main advantage is not just benchmark performance. It is long-horizon execution.
Anthropic says Fable 5 can work autonomously longer than previous Claude models and stay focused across millions of tokens in long-running tasks. That matters for agentic workflows because many failures happen when a model loses track of:
- the original goal,
- previous decisions,
- file-level context,
- intermediate plans,
- or constraints introduced earlier in the run.
A useful example from Anthropic is the Slay the Spire test. When Fable 5 was given persistent file memory, so it could write and read notes across the run, it performed 3x better than Claude Opus 4.8. That suggests Fable 5 is not just handling more text; it is using long-running memory more effectively.
Good fit use cases include:
1. Large engineering tasks
Use Fable 5 for tasks that require multiple steps across a codebase:
- refactors,
- framework migrations,
- test generation,
- dependency upgrades,
- API surface analysis,
- code review across many files,
- agentic debugging.
If you already use Claude Code-style workflows, patterns from dynamic Claude Code workflows on Opus 4.8 carry over well. Fable 5 is better suited when the run is longer and context retention is the bottleneck.
2. Deep document analysis
Fable 5 is useful when you need to reason across long or dense material:
- financial documents,
- contracts,
- research papers,
- technical specs,
- internal docs,
- tables and charts.
3. Vision and multimodal analysis
Fable 5 includes vision capability, so you can use it for workflows that combine text and images.
4. Scientific and technical reasoning
Anthropic reports strong performance on scientific research tasks. For public users, this runs with safeguards enabled.
Benchmarks and real-world results
Anthropic describes Fable 5 as state-of-the-art on nearly all tested benchmarks. The named benchmark results include:
| Benchmark | Reported result | Why developers should care |
|---|---|---|
| FrontierCode by Cognition | Highest among frontier models at medium effort | Measures realistic coding tasks |
| Finance Benchmark by Hebbia | Highest score | Useful signal for document reasoning, charts, and tables |
| CursorBench | State-of-the-art | Relevant to coding agents and long-horizon IDE workflows |
| FrontierBench by Cognition | Highest-scoring model | Emphasizes long-horizon reasoning |
The most concrete real-world example is from Stripe. Stripe used Fable 5 to migrate a 50-million-line Ruby codebase in one day. The team estimated the same work would have taken over two months manually.
That is the kind of workload where Fable 5 makes the most sense: expensive, multi-step work where the model must keep context across a large codebase and a long execution path.
You can read Anthropic’s original claims in the official Claude Fable 5 and Mythos 5 announcement.
How Fable 5 safety safeguards work
Fable 5 is not simply Mythos 5 exposed to the public. It includes safeguards around sensitive areas.
Instead of always refusing protected queries outright, Anthropic says certain requests are routed or fall back to Claude Opus 4.8. That means a user may still receive an answer, but the request is not handled with Fable 5’s full frontier capability for that protected area.
Anthropic says these safeguards trigger in less than 5% of sessions on average.
Protected areas include:
-
Cybersecurity
- exploit development,
- offensive cyber tasks,
- agentic hacking.
-
Biology and chemistry
- sensitive biological or chemical queries,
- AAV design,
- bioweapons-related requests.
-
Distillation
- attempts to extract model behavior to train competing models.
Implementation note: if your app operates near any of these domains, test fallback behavior. You should not assume every request will be answered by Fable 5 itself.
For normal coding, documentation, support, analysis, and product workflows, this routing should usually be invisible.
Pricing
Claude Fable 5 pricing:
| Model | Input | Output |
|---|---|---|
| Claude Fable 5 | $10 / 1M tokens | $50 / 1M tokens |
| Claude Mythos 5 | $10 / 1M tokens | $50 / 1M tokens |
| Claude Opus 4.8 | $5 / 1M tokens | $25 / 1M tokens |
Fable 5 costs roughly 2x Claude Opus 4.8 per token. Use it when the value of better long-horizon reasoning exceeds the additional token cost.
For cost planning, read the Claude Fable 5 pricing guide. If you are comparing against the lower tier, see the Claude Opus 4.8 pricing breakdown. Anthropic also keeps current numbers on its pricing page.
Claude Fable 5 vs Claude Opus 4.8
For most teams, the real choice is not Fable 5 vs Mythos 5. It is Fable 5 vs Opus 4.8.
| Aspect | Claude Fable 5 | Claude Opus 4.8 |
|---|---|---|
| Positioning | Premium frontier tier | High-capability tier below Fable |
| Input price | $10 / 1M tokens | $5 / 1M tokens |
| Output price | $50 / 1M tokens | $25 / 1M tokens |
| Context behavior | Built for long-running tasks across millions of tokens | 1M-token context window |
| Long-horizon autonomy | Stronger; 3x better in the Slay the Spire persistent-memory test | Strong, but below Fable 5 |
| Safety role | Routes sensitive queries down to Opus 4.8 | Acts as fallback for protected Fable 5 requests |
Use this rule of thumb:
- Use Fable 5 for long, complex, multi-step work.
- Use Opus 4.8 for shorter, routine, or cost-sensitive tasks.
Good Fable 5 workloads:
- large codebase migration,
- long agent runs,
- deep document synthesis,
- complex debugging,
- multi-file refactoring,
- extended research workflows.
Good Opus 4.8 workloads:
- simple code generation,
- short Q&A,
- summarization,
- routine support tasks,
- lightweight automation.
For broader model comparison context, see Claude Opus 4.8 vs GPT-5.5 vs Gemini 3.5. If budget is the main blocker, review the free Opus 4.8 access options before committing spend.
How to call Claude Fable 5 from the API
At a high level, your request needs:
- Anthropic API key,
- Messages API endpoint,
-
model: "claude-fable-5", - max token limit,
- message array.
Example request shape:
curl https://api.anthropic.com/v1/messages \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{
"model": "claude-fable-5",
"max_tokens": 1024,
"messages": [
{
"role": "user",
"content": "Review this migration plan and identify implementation risks."
}
]
}'
When testing, inspect the response usage field because it directly maps to cost:
{
"usage": {
"input_tokens": 1200,
"output_tokens": 650
}
}
A simple cost estimate:
input_cost = input_tokens / 1_000_000 * 10
output_cost = output_tokens / 1_000_000 * 50
total_cost = input_cost + output_cost
For production, wrap this in logging so you can track cost per feature, user, job, or agent run.
Testing the Claude Fable 5 API with Apidog
Before wiring Fable 5 into production code, test it with an API client. Apidog gives you a workspace for designing, sending, saving, and debugging API requests.
A practical setup flow:
1. Create a new request
Create a request for the Anthropic Messages API endpoint.
Set the method to:
POST
Add the request URL for the Messages API.
2. Add authentication
Store your Anthropic API key as an environment variable instead of hardcoding it into the request body or scripts.
Example variable:
ANTHROPIC_API_KEY
Then reference it in the request headers.
3. Add required headers
Use headers similar to:
x-api-key: {{ANTHROPIC_API_KEY}}
anthropic-version: 2023-06-01
content-type: application/json
4. Set the model
In the JSON body, use:
{
"model": "claude-fable-5"
}
5. Save reusable request templates
Create saved requests for common tasks:
- code review,
- migration planning,
- document analysis,
- prompt regression tests,
- streaming response tests.
That gives your team a shared baseline for testing prompts and parameters.
6. Track usage
Check token usage on every response while experimenting. Fable 5 output is more expensive than input, so verbose responses can change your cost profile quickly.
Apidog is useful here because you can:
- reuse auth safely,
- test streaming responses,
- inspect response payloads,
- compare prompt versions,
- save repeatable requests,
- share request collections with teammates.
If you have already followed the guide on testing the ChatGPT API with Apidog, the workflow is similar. You mainly change the base URL, headers, and model field for Anthropic.
For deeper request-body examples, see the Claude Fable 5 API guide. You can also Download Apidog and test requests directly.
Implementation checklist
Before using Fable 5 in production, validate these items:
- [ ] Confirm the model ID is
claude-fable-5. - [ ] Log
input_tokensandoutput_tokens. - [ ] Estimate cost per workflow, not just per request.
- [ ] Test long-running prompts with real production-like context.
- [ ] Compare the same tasks against Opus 4.8.
- [ ] Decide when Fable 5 is worth the 2x token cost.
- [ ] Handle possible safeguard routing for sensitive topics.
- [ ] Add retries and timeout handling for long-running requests.
- [ ] Use streaming where long outputs are expected.
- [ ] Save prompt and parameter versions for regression testing.
FAQ
Is Claude Fable 5 free?
No. Claude Fable 5 is a paid frontier model priced at $10 per million input tokens and $50 per million output tokens through the Claude API and consumption-based Enterprise plans.
There is no free tier for Fable 5 itself. If you want no-cost options for Anthropic models, see the Opus 4.8 no-cost access paths. A separate free Fable 5 access guide covers any trial-style options for the new model.
What does “Mythos-class” mean?
“Mythos-class” describes a capability tier, not a separate architecture. Fable 5 and Mythos 5 share the same underlying model. Fable 5 adds safeguards so the model can be released for general use.
Can I get Claude Mythos 5?
Not through normal API access. Mythos 5 is restricted. It is deployed through Project Glasswing to a small group of cyberdefenders and infrastructure providers, with some cyber safeguards lifted. Biology researchers are expected to get a version later with biology and chemistry safeguards removed. Anthropic has also said a broader trusted-access program is planned.
Is Fable 5 better than Opus 4.8?
On raw capability, yes. Fable 5 is the more capable frontier model and showed a 3x improvement over Opus 4.8 in the Slay the Spire persistent-memory test.
But Opus 4.8 costs half as much, at $5 input and $25 output per million tokens. For shorter or routine tasks, Opus 4.8 may be the better value. For more context on that tier, see what Claude Opus 4.8 is.
What is the Fable 5 API model name?
The model ID is:
claude-fable-5
Pass that string in the model field of your Anthropic Messages API request.
Why did my Fable 5 response feel different?
It may have hit a safeguard. Queries related to cybersecurity, biology and chemistry, or model distillation can be routed to Claude Opus 4.8 instead of being answered directly by Fable 5. Anthropic says this happens in less than 5% of sessions on average.
Does Fable 5 have a bigger context window than Opus 4.8?
Fable 5 is built for long-running tasks and can stay focused across millions of tokens. Opus 4.8 has a 1M-token context window. Both support long-context workflows, but Fable 5’s advantage is strongest in sustained, multi-step work.
How much can Fable 5 do in one task?
Anthropic’s strongest example is Stripe using Fable 5 to migrate a 50-million-line Ruby codebase in one day. Stripe estimated the work would have taken more than two months manually.
Final take
Claude Fable 5 is Anthropic’s new public frontier model: Mythos-class capability with safeguards, priced at $10 input and $50 output per million tokens, and designed for long-horizon work.
Use it when the task is large enough that context retention, autonomy, and multi-step reasoning matter more than raw token cost. For routine work, compare it against Opus 4.8 before standardizing.
The fastest way to evaluate it is to send real workload-shaped requests, inspect the outputs, and track token usage. Download Apidog, create a request with claude-fable-5, and test it against your own codebase, documents, or agent workflow.
Top comments (0)