Back to HomeAI API

Claude Fable 5.2 Guide (2026): API Pricing, Differences from Fable 5.1, Mythos 5.2, and Migration

17 min min read
#Claude Fable 5.2#Claude Fable 5.1#Claude Mythos 5.2#Claude Opus 5.5#Claude API#Anthropic#AI API#API Pricing#Model Migration#Enterprise Procurement

Illustration of a price cut on Anthropic's new Fable model: two price boards side by side, old and new, with every fee lower on the new one, next to a door that opens only for verified organizations, representing the Mythos version with the same specs

What Is Claude Fable 5.2: The New Top Model, Released November 9

Anthropic's model page states that Claude Fable 5.2 was released on November 9, 2026, "for demanding reasoning and long-horizon agentic work." Typical uses Anthropic lists are agent sessions that run for hours, multistep deep research, and analysis carried through to a finished document, spreadsheet, or deck. It replaces Fable 5.1 as the top model in the current lineup; Fable 5.1 moves to the list of older models but can still be called (lifecycle status Active).

Current Lineup and Pricing

ModelModel IDInput/Output (per million tokens)Cache readContextMax outputDefault effort
Claude Fable 5.2claude-fable-5-2$8/$40$0.201M128Khigh
Claude Opus 5.5claude-opus-5-5$4/$20$0.201M128Kmedium
Claude Sonnet 5.5claude-sonnet-5-5$2/$10$0.101M128Khigh
Claude Haiku 5.5claude-haiku-5-5From $0.10/$0.50 ($0.50/$2.50 for prompts over 100K)From $0.01 ($0.05 over 100K)1M128Kmedium

Sources: Anthropic's official Fable 5.2 model page, What's new page, and pricing page (checked 2026-11-19), USD. Fable 5.1 is an older model (still available) at $10/$50.

Pricing: Every Item 20% Below Fable 5.1

Anthropic's What's new page says input and output cost less on Fable 5.2 than on Fable 5.1, and cache reads stay at 0.025 times the base input price, the same multiplier as Fable 5.1. Dividing the two sets of official prices item by item, every item comes out exactly 20% lower. That 20% is arithmetic on two official prices, not a figure Anthropic advertises.

Item (per million tokens)Fable 5.2Fable 5.1Opus 5.5
Input$8$10$4
Output$40$50$20
5-minute cache write$10$12.50$5
1-hour cache write$16$20$8
Cache read$0.20$0.25$0.20
Batch API (input/output)$4/$20$5/$25$2/$10
Fast mode (input/output)$16/$80Not on the supported list$8/$40

A few things that are easy to miss when reading the price list:

  1. No surcharge for the 1M-token context: Anthropic's pricing page says Claude 4.6 and later models (except Haiku 5.5) include the full 1M context at standard pricing, so a 900K-token request is billed at the same per-token rate as a 9K-token request; Fable 5.2 is one of them
  2. Cache reads cost $0.20, the same as Opus 5.5: Fable 5.2's input price is twice Opus 5.5's, but its cache-read multiplier is lower (0.025x versus 0.05x), so both come to the same price
  3. Fast mode is priced separately: Anthropic says fast mode provides up to 2.5x higher output tokens per second at premium pricing; for Fable 5.2 on the Claude API it costs $16/$80, and it can't be combined with the Batch API
  4. US-only inference costs more: on Claude 4.6 and later models, inference_geo: "us" applies a 1.1x multiplier to all token categories; the default, global, uses standard pricing
  5. The tokenizer hasn't changed: Fable 5.2 uses the tokenizer introduced with Claude Opus 4.7, and the pricing page says Claude 4.7 and later models all use it, so moving from Fable 5.1 or Opus 5.5 doesn't add tokens to the same text because of the tokenizer

Mythos 5.2: Same Specs and Price, Verified Organizations Only

Claude Mythos 5.2 (claude-mythos-5-2) launched the same day as Fable 5.2. Anthropic says Mythos 5.2 offers the same capabilities only to organizations verified through Anthropic's verification programs, such as the Cyber Verification Program, and shares Fable 5.2's specifications and pricing ($8/$40, $0.20 cache reads). To request access, apply to the program that covers your use case, or contact your Anthropic account team.

For most enterprises, Fable 5.2 is the one you can use directly: the migration guide says Fable 5.2 doesn't require access approval on the Claude API. There is one more difference: Fable 5.2 has its own rate limit, while Mythos 5.2 shares one combined limit with Mythos 5.1 and Mythos 5. For background on the Mythos tier, see What Is the Mythos Model?.

Key Specs: How It Differs from Fable 5.1 and Opus 5.5

ItemFable 5.2Fable 5.1Opus 5.5
Context / max output1M / 128K1M / 128K1M / 128K
ThinkingAdaptive, always on, can't be turned offAdaptive, always onAdaptive, always on
Default efforthighhighmedium
Forced tool use (any, named tool)Returns 400Returns 400Returns 400
Assistant prefillReturns 400Returns 400—
temperature, top_p, top_kOmit, per AnthropicOmit, per Anthropic—
Data retention30-day retention required; not available under ZDR30-day retention requiredAvailable under ZDR
Older computer use tool computer_20251124Returns 400 (except on Amazon Bedrock)Accepted (with beta header)—
Structured outputs in CMEK organizationsNot supportedNot supportedSupported
Priority TierNot supported——
Rate limitIts ownShared with Fable 5Separate from Fable 5.2, with higher token limits at each tier

Other official specs: text and image input, text output; reliable knowledge cutoff June 2026; five effort levels, low, medium, high, xhigh, and max; retirement not sooner than 2027-11-09. Available on the Claude API, Amazon Bedrock (global.anthropic.claude-fable-5-2 or us.anthropic.claude-fable-5-2, on the bedrock-runtime endpoint only), Claude Platform on AWS, Google Cloud (Google managed and Anthropic managed), and Microsoft Foundry (on Anthropic infrastructure).

Anthropic's guidance on effort: start at the default, high, because most of the gain in capability is already present there; step up to xhigh for your hardest long-running coding and agentic work where you've measured a gain; reserve max for measuring the model's ceiling and the rare task where time and tokens don't matter.

Cost Math (Arithmetic, Not Benchmarks)

Assumptions: 100 million input tokens and 10 million output tokens per month, with the same token volume on every model (all three use the same tokenizer). In practice Fable 5.2 defaults to high effort and Opus 5.5 to medium; the migration guide warns that a request that omits effort runs one level higher on Fable 5.2, which shows up as different cost, latency, and results, so this comparison is unit-price arithmetic only. USD, standard billing, excluding data residency and other modifiers.

Scenario A: No Caching

MethodInput costOutput costMonthly total
Fable 5.1100M × $10 = $1,00010M × $50 = $500$1,500
Fable 5.2100M × $8 = $80010M × $40 = $400$1,200
Fable 5.2 on the Batch API (offline work)100M × $4 = $40010M × $20 = $200$600
Fable 5.2 with fast mode100M × $16 = $1,60010M × $80 = $800$2,400
(Reference) Opus 5.5100M × $4 = $40010M × $20 = $200$600

At the same volume, Fable 5.2 saves $300 a month (20%) over Fable 5.1; fast mode costs twice the standard price.

Scenario B: Caching as Agent Workloads Typically Use It

The same 100 million input tokens per month, of which 80 million are cache reads, 5 million are 5-minute cache writes, and 15 million are uncached; 10 million output tokens.

MethodCache readsCache writesUncached inputOutputMonthly total
Fable 5.180M × $0.25 = $205M × $12.50 = $62.5015M × $10 = $150$500$732.50
Fable 5.280M × $0.20 = $165M × $10 = $5015M × $8 = $120$400$586
(Reference) Opus 5.580M × $0.20 = $165M × $5 = $2515M × $4 = $60$200$301

With caching, Fable 5.2 drops from $1,200 to $586, about 51% less, but the $400 of output is about 68% of the bill. In other words, once caching has pushed input costs down, output volume is what drives a Fable 5.2 bill, and higher effort means more thinking and more output. At the same volume, Fable 5.2 costs about 1.95 times as much as Opus 5.5.

⚠️ Scenario B assumes a high cache hit rate. When requests are more than 5 minutes apart, the cache expires and has to be written again, and the write price ($10) is higher than uncached input ($8); for low-frequency workloads, estimate your hit rate before turning caching on. Anthropic also notes that at higher effort levels such as xhigh and max, Fable 5.2 can start subagents to review its own work, and subagents can account for a large part of what a session costs.

Migrating from Fable 5.1: Four Changes That Break Code

Anthropic calls the migration "mostly drop-in": the API surface, always-on adaptive thinking, the forced tool choice restriction, refusal handling, and the 30-day data retention requirement all match Fable 5.1. First change the model ID from claude-fable-5-1 to claude-fable-5-2, then check these four:

  1. Editing earlier turns returns 400: every Fable 5.2 request gets a prefix check. If you change system, tools, or earlier messages mid-conversation and then send thinking blocks back, you get a 400 error with the message The block is bound to a different conversation, and retrying the same request body doesn't clear it. Nothing changes if Claude Code, claude.ai, Claude Managed Agents, or the Claude Agent SDK builds your requests. If your code builds messages itself, keep the conversation append-only; as a stopgap while you fix it, send the thinking-binding-controls-2026-08-01 beta header and set thinking.block_binding.prefix_mismatch_behavior to "drop_block", though Anthropic warns that dropping blocks can raise output tokens
  2. Fable 5.1 can't read Fable 5.2 thinking blocks: moving from Fable 5.1 to Fable 5.2 keeps the reasoning; moving back to Fable 5.1, Opus 5.5, or another model makes the API drop Fable 5.2's thinking blocks (the request succeeds and dropped blocks aren't billed), which can raise cost and latency on the first turn after the switch
  3. The older computer use tool returns 400: computer_20251124 returns 400 on every platform except Amazon Bedrock; declare computer_toolset_20260801 instead
  4. The advisor tool needs a matching model: with Fable 5.2 as the executor, the advisor must be Fable 5.2 or Mythos 5.2; a Fable 5.1 advisor returns 400

These don't return errors, but behavior changes:

  • Prompt lines asking for less thinking no longer work: Fable 5.1 usually follows a line such as "answer directly without deliberating"; Fable 5.2 still thinks at medium effort and above, and at high and above it thinks before almost every reply. To save time or cost, use a lower effort level
  • A small max_tokens can return thinking and no text: thinking tokens count toward max_tokens, so a request that relied on a "don't think" line to fit a small limit can stop with stop_reason: "max_tokens" on Fable 5.2 before any text
  • A turn can end before the work is done: on long tasks the model sometimes lists what remains or offers to continue, more often at low and medium effort
  • Separate rate limits: Fable 5.2 has its own rate limit, separate from the one Fable 5.1 and Fable 5 share; check it before moving production traffic

Moving Up from Opus 5.5: Check Data Retention and Effort First

Fable 5.2 uses the same Messages API and tool use patterns as Opus 5.5. Both always think adaptively and both return 400 for forced tool use, so those parts carry over. A few things need attention first:

  1. ZDR organizations must enable 30-day retention: an organization running Opus 5.5 under a zero data retention arrangement gets a 400 invalid_request_error on every Claude API request to Fable 5.2 until it enables 30-day retention or confirms eligibility with Anthropic
  2. Default effort is one level higher: Opus 5.5 defaults to medium and Fable 5.2 to high; a request that omits effort still succeeds, but cost, latency, and results change. Anthropic recommends setting effort explicitly and re-running an effort sweep on your own evals
  3. No structured outputs in CMEK organizations: organizations with customer-managed encryption keys should remove strict: true and output_config.format and validate output in their own code
  4. Lower rate limits: Fable 5.2's rate limit is separate from Opus 5.5's, with lower token limits at each tier
  5. Refusals and fallback: Fable 5.2 runs safety classifiers, so handle every stop_reason: "refusal" category; the permitted fallback targets are Opus 4.8 and Opus 5

Instead of forced tool use, keep tool_choice at {"type": "auto"}, name the tool in your instruction, and set strict: true on the tool definition, or use JSON outputs. A request that sets effort explicitly:

{
  "model": "claude-fable-5-2",
  "max_tokens": 32000,
  "output_config": { "effort": "high" },
  "tool_choice": { "type": "auto" },
  "messages": [{ "role": "user", "content": "..." }]
}

You can omit thinking or send {"type": "adaptive"}; {"type": "disabled"} and {"type": "enabled"} with budget_tokens both return 400.

CloudInsight Enterprise Claude API Procurement | Local Invoices, No Foreign Credit Card Needed

Fable 5.2 is cheaper, but a top-tier model's bill still runs into thousands of dollars. What enterprises get stuck on most is often not the technology but payment, invoicing, and budget control.

Contact CloudInsight for a Claude API Enterprise Quote

We offer one-stop procurement for Claude, GPT, and Gemini, Taiwan uniform invoices, and technical support in Chinese.

How to Choose: Fable 5.2, Fable 5.1, or Opus 5.5

WorkloadRecommendationWhy
Most workloads, or unsure where to startOpus 5.5Anthropic recommends starting most workloads on the latest Opus; half Fable 5.2's unit price
Demanding reasoning or long-horizon agentic work where Opus 5.5 at higher effort still misses your evalsFable 5.2Anthropic's stated condition for moving up; 20% cheaper than Fable 5.1
Currently on Fable 5.1Move to Fable 5.2Same API surface, every price 20% lower; handle the four 400-error changes above first
Organization under a zero data retention (ZDR) arrangementStay on Opus 5.5 for nowFable 5.2 requires 30-day retention and returns 400 without authorization
Organizations verified through programs such as the Cyber Verification ProgramConsider Mythos 5.2Same specs and price as Fable 5.2

Fable 5.1 is still Active and can be called normally. Code that hasn't handled the changes above can stay on Fable 5.1 for now, but the same volume costs about 25% more per month ($10 versus $8).

FAQ: Claude Fable 5.2

How much does Claude Fable 5.2 cost?

$8 per million input tokens and $40 per million output tokens; cache reads $0.20, 5-minute cache writes $10, and 1-hour cache writes $16. The Batch API is $4/$20, and fast mode is $16/$80. The full 1M-token context is billed at standard pricing.

How much cheaper is Fable 5.2 than Fable 5.1?

Input, output, cache reads, both cache write types, and batch prices are each 20% below Fable 5.1. At 100 million input and 10 million output tokens a month, Fable 5.1 costs $1,500 and Fable 5.2 costs $1,200.

Can we still use Fable 5.1?

Yes. Fable 5.1 has moved to the list of older models, but its lifecycle status is still Active and its price stays at $10/$50.

How is Mythos 5.2 different from Fable 5.2?

Same specs and price; the difference is eligibility. Mythos 5.2 is available only to organizations verified through Anthropic's verification programs, such as the Cyber Verification Program, while Fable 5.2 needs no access approval on the Claude API.

What code changes does Fable 5.2 need?

From Fable 5.1: change the model ID, keep conversations append-only, move computer use to computer_toolset_20260801, and set the advisor to Fable 5.2 or Mythos 5.2. From Opus 5.5: confirm 30-day data retention first and set effort explicitly.

Can Fable 5.2 turn off thinking or force a tool call?

Neither. Adaptive thinking is always on; sending disabled or budget_tokens returns 400, so effort is the only control for depth. Setting tool_choice to any or a named tool also returns 400; use auto with strict tools or JSON outputs instead.

Conclusion: Fable 5.1 Users Should Switch; Everyone Else Starts on Opus 5.5

Three Core Recommendations

  1. If you're on Fable 5.1, plan the move to Fable 5.2: every price is 20% lower and the API surface is the same; run the prefix check test before going live
  2. If you're not on Fable yet, evaluate Opus 5.5 first: that's Anthropic's recommendation; move to Fable 5.2 only if Opus 5.5 at higher effort still isn't enough
  3. Budget around output volume and effort: once caching has pushed input down, output is most of the bill; set effort explicitly and re-baseline cost and latency

Ready to Adopt Claude Fable 5.2?

Contact the CloudInsight sales team and we'll help plan your Fable 5.2 and Opus 5.5 usage mix and handle payment and invoicing.

Join our LINE Official Account for instant model-selection consultation.


Further Reading

References

Need Professional Cloud Advice?

Whether you're evaluating cloud platforms, optimizing existing architecture, or looking for cost-saving solutions, we can help

Book Free Consultation

Related Articles

AI API

Claude Opus 5.5 and Sonnet 5.5: API Pricing, Migration Changes, and Model Choice

Anthropic released Claude Opus 5.5 (Sept 22) and Sonnet 5.5 (Sept 28) in September 2026. Opus 5.5 is priced at $4/$20 — 20% below Opus 5 — and is now Anthropic's official default for most workloads; Sonnet 5.5 stays at $2/$10. This guide covers official specs, cache pricing, cost math, and the API changes that break code migrating from Opus 5 or Sonnet 5.

AI API

Claude Haiku 5.5 Guide: Tiered API Pricing, Differences from Haiku 4.5, and Migration

Anthropic released Claude Haiku 5.5 on 2026-10-07. For prompts up to 100,000 tokens it costs just $0.10/$0.50 per million tokens, one tenth of Haiku 4.5. But a request whose prompt is over 100,000 tokens is billed at $0.50/$2.50 in full, and the new tokenizer counts the same text as roughly 30% more tokens. This guide covers official specs, both price tiers, cost math that accounts for both traps, and the changes that break or alter behavior when migrating from Haiku 4.5.

AI API

Gemini 2.5 Pro and Flash Auto-Route From December 1: Which Model You Get, Pricing, and the Code to Change (2026)

On 2026-11-02 Google announced that starting 2026-12-01, every API request to gemini-2.5-pro will automatically route to gemini-3.8-flash and every request to gemini-2.5-flash to gemini-3.5-flash-lite, billed at the new models' rates. This is not a shutdown, and requests won't fail because of the model name, but function calling validation, thinking settings, sampling parameters, and prefilled turns all have breaking changes. This guide covers the routing map, prices before and after, cost math, and the code to change before go-live.