Claude Fable 5.1 API: Pricing, What's New, and How to Use It

Anthropic released Claude Fable 5.1 on September 1, 2026. Positioned by the company among "the world's most advanced models" for coding and knowledge work, Fable 5.1 keeps Fable 5's 1M-token context and 128K-token output; the real difference shows up as effort increases, in agentic coding, and in long-running work. There is a notable cybersecurity change too: fewer false alarms, vulnerability discovery yes, exploit development no.

This guide covers what Fable 5.1 adds on top of Fable 5, why the cache-read discount Anthropic announced does not carry over to your Onysoft bill, which jobs belong to Fable 5.1 versus Opus 5 and Sonnet 5, and how to move an existing Fable 5 integration over with a one-line change. Every example runs against Onysoft AI Gateway's OpenAI-compatible API, so if you already use the OpenAI SDK, you are most of the way there.

What Claude Fable 5.1 Adds on Top of Fable 5

Fable 5.1 is not a new family; it is the direct successor to Fable 5. The technical frame is unchanged: a 1M-token context, a 128K-token maximum output, and the anthropic/claude-fable-5.1 identifier on Onysoft. According to Anthropic, the gains fall into three areas:

• Performance that scales with effort: at low and medium effort it matches or beats Fable 5; at high effort the gap widens considerably. In practice, migrating from Fable 5 should not cost you anything on simple jobs, and the payoff grows on hard jobs where the model is allowed to think longer.
• Agentic coding: multi-file changes, long code refactors, and tool-calling agent loops are the focus of this release. Per results published in the press (MarkTechPost), Fable 5.1 scored 52.6% on Terminal-Bench-Science.
• Long-running knowledge work: multi-step analyses and long agent workflows. The fastest way to verify this on your own workload is to run the same task side by side with Fable 5 and Fable 5.1 in the Playground.

If you want to control effort through the API, Onysoft forwards the reasoning parameter (e.g. {"effort": "high"}) as-is to the provider for models that support it; you will find an example in the code section below. Keep in mind that higher effort can mean more tokens consumed, which shows up in cost.

Cybersecurity: Fewer False Alarms, Vulnerability Discovery Yes, Exploits No

One of the biggest problems with model-driven code scanning is the false-alarm load that eats into a team's time. According to Anthropic, Fable 5.1 produces roughly 60% fewer false positives on cybersecurity findings in Claude Code. In audit workflows where every bogus finding needs an engineer to review it, that translates directly into less work.

The boundary is clearly drawn as well: Fable 5.1 can discover security vulnerabilities in a codebase, but it does not develop exploits that weaponize them. That distinction defines the use cases:

• A good fit: code review, dependency and configuration audits, prioritizing findings, remediation suggestions, and patch drafts.
• Not a fit: expecting working attack code or proof-of-concept exploits; the model is not designed for that.

A note on Mythos 5.1: alongside Fable 5.1, Anthropic also announced Mythos 5.1, which is available by invitation only. Mythos 5.1 (and Mythos 5) are not in the Onysoft catalog; the most recent Fable model you can reach through Onysoft is anthropic/claude-fable-5.1.

Fable 5.1 Pricing and the Cache Discount: How Onysoft Bills It

On Onysoft, Fable 5.1 is offered at the same price as Fable 5. Sale prices for the relevant Claude models:

Model	API identifier	Context	1M input	1M output	1M input (TL)	1M output (TL)
Claude Fable 5.1	anthropic/claude-fable-5.1	1M	$15.00	$75.00	727.41 TL	3,637.06 TL
Claude Fable 5	anthropic/claude-fable-5	1M	$15.00	$75.00	727.41 TL	3,637.06 TL
Claude Opus 5	anthropic/claude-opus-5	1M	$7.50	$37.50	363.71 TL	1,818.53 TL
Claude Sonnet 5	anthropic/claude-sonnet-5	1M	$3.00	$15.00	145.48 TL	727.41 TL
Claude Haiku 4.5	anthropic/claude-haiku-4.5	200K	$1.50	$7.50	72.74 TL	363.71 TL

Prices are Onysoft sale prices as of September 13, 2026; see /models for current pricing. TL equivalents use 1 USD = 48.4941 TL (Central Bank of the Republic of Türkiye, EVDS rate for September 11, 2026).

An honest note on the cache-read discount: with Fable 5.1, Anthropic announced a 75% cut to its cache-read price; according to the provider, that means savings of about 25% on typical workloads and up to 45% on heavy agentic work. This discount applies to the cache-read price on Anthropic's direct API; when you use the model through Onysoft, the standard input/output sale price applies. Onysoft bills every request as the model's input and output token counts multiplied by the sale prices above, and there is no separate cache tier in our price table. Do not factor this discount into estimates of your Onysoft bill.

The real cost levers on the Onysoft side are these: send repeated long context only when it is needed, cap output with max_tokens, reserve high effort for genuinely hard steps, and spread high-volume traffic to Opus 5 or Sonnet 5 using the decision matrix below. For a deeper look at tiered model strategies, see the AI API cost optimization guide.

Fable 5.1 vs Opus 5 vs Sonnet 5: Which Model for Which Job?

The price gap is clear: per token, Fable 5.1 costs 2x Opus 5 and 5x Sonnet 5. So the right question is not "which one is stronger" but "which one does this job belong to":

Workload	Recommended model	Why
Long-running agentic coding, multi-file refactors	anthropic/claude-fable-5.1	The focus of this release; per the provider, the biggest gain over Fable 5 comes at high effort.
Vulnerability scanning and triage across a codebase	anthropic/claude-fable-5.1	Fewer false positives per the provider; discovers vulnerabilities, does not develop exploits.
Deep reasoning, long document review, architecture analysis	anthropic/claude-opus-5	Same context (1M) and output (128K) limits as Fable 5.1, at half the token price.
Everyday production traffic, coding assistance, RAG, agent subtasks	anthropic/claude-sonnet-5	Price/performance balance; most of your traffic likely belongs here.
Classification, short summaries, high-volume routine calls	anthropic/claude-haiku-4.5	Lowest unit cost in the table; 200K context is plenty for these jobs.

When should you not use Fable 5.1?

• For short Q&A, simple extraction, and templated text: you pay 5x Sonnet 5 per output token, and the difference rarely shows in the result.
• For one-off deep analysis where Opus 5 is enough: it runs with the same context and output limits at half the price.
• For high-volume pipelines (labeling, classification): Haiku 4.5 or Sonnet 5 is far more economical.
• For scenarios where you expect exploit or attack code: the model is not designed for that.

A practical rule: build the prototype on Sonnet 5, escalate cases where quality falls short to Opus 5, and reserve Fable 5.1 for long agent loops and critical coding steps. For Opus 5 details, see the Claude Opus 5 API guide; for the whole family, see the Claude models page. The same key reaches all 754+ models in the catalog.

Step-by-Step Access (with TL Billing in Türkiye) and Migrating from Fable 5

You do not need a separate Anthropic contract to use Fable 5.1; Onysoft AI Gateway gets you started in a few minutes:

1. Create a free account — corporate e-invoicing and KVKK compliance are standard.
2. Generate an API key with the sk-ony- prefix from the dashboard.
3. Add balance; usage is deducted pay-as-you-go, in TL at the central bank (TCMB) rate. No subscription, no monthly commitment.
4. Try anthropic/claude-fable-5.1 in the Playground before writing any code.
5. Set your SDK's base_url to https://api.onysoft.com/v1 and pass the model name.

Migrating from Fable 5: the only thing that changes in your code is the model name: anthropic/claude-fable-5.1 instead of anthropic/claude-fable-5. Price, context, and output limits are the same. anthropic/claude-fable-5 remains active in the catalog, so you can shift traffic gradually and compare both versions on your own tasks.

If you would rather not update your code for every new Fable release, use the ~anthropic/claude-fable-latest alias (the leading ~ is required); it points to the most recent Fable version. If you need reproducible, predictable behavior in production, calling the pinned identifier (anthropic/claude-fable-5.1) is the safer choice. For the wider local picture, see the Claude API in Türkiye guide; for the previous release, see the Claude Fable 5 API guide. Support is available 24/7 if you run into trouble.

First Request with Python, curl, and Streaming

Because the Onysoft endpoint follows the OpenAI schema, the official openai Python package works directly. Streaming responses arrive as raw OpenAI chunks, so they are fully SDK-compatible, and streaming is the recommended approach for long coding and agent tasks anyway:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.onysoft.com/v1",
    api_key="sk-ony-YOUR_KEY",
)

stream = client.chat.completions.create(
    model="anthropic/claude-fable-5.1",
    messages=[
        {"role": "user", "content": "Find the security vulnerabilities in this module, rank them by severity, and suggest a fix for each."}
    ],
    stream=True,
)
for chunk in stream:
    if not chunk.choices:
        continue
    delta = chunk.choices[0].delta.content
    if delta:
        print(delta, end="")

A non-stream request to the same model with curl:

curl https://api.onysoft.com/v1/chat/completions \
  -H "Authorization: Bearer sk-ony-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-fable-5.1",
    "messages": [{"role": "user", "content": "Assess the risks of this refactoring plan step by step."}]
  }'

Non-stream responses are wrapped in a success/data envelope; the OpenAI-format response object is in the data field. For streaming, add "stream": true to the body, and use curl's -N flag so chunks print as they arrive:

curl -N https://api.onysoft.com/v1/chat/completions \
  -H "Authorization: Bearer sk-ony-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-fable-5.1",
    "stream": true,
    "messages": [{"role": "user", "content": "Draft a phased migration plan for this codebase."}]
  }'

To set the effort level, send the reasoning parameter through extra_body in the Python SDK; it is forwarded as-is to the provider for models that support it:

stream = client.chat.completions.create(
    model="anthropic/claude-fable-5.1",
    messages=[{"role": "user", "content": "Prepare a detailed plan for splitting this service into modules."}],
    extra_body={"reasoning": {"effort": "high"}},
    stream=True,
)

OnyRouter tip: for mixed traffic where not every request needs Fable 5.1, set the model to onysoft/auto and OnyRouter picks a suitable model for each request. Routing is free; you pay only the price of the selected model. For steps where Fable 5.1 must run, pin the model name. See the OnyRouter guide for details, and the API documentation for supported request parameters and error codes.