Claude Fable 5.2 Limits
A complete technical reference to context capacity, developer API rate tiers, monthly spend caps, and subscription plan limits for Claude Fable 5.2.
Claude Fable 5.2 limits feature a 1 million-token context window and a maximum output capacity of 128,000 tokens. Dedicated API rate limit tiers start at 4,000 requests per minute (RPM), and monthly account spend ceilings range from $500 to $200,000.
Released by Anthropic on November 9, 2026, Claude Fable 5.2 handles high-volume reasoning and coding tasks under a dedicated rate-limiting bucket that operates independently from earlier models like Claude Fable 5.1. While developer API access is governed by structured throughput tiers and uncached token metering, individual web and desktop plans restrict Fable model family activity to a maximum of 50% of total weekly usage allowances.
Understanding these operational thresholds allows engineering and procurement teams to architect stable agentic pipelines without triggering service pauses or unexpected budget cutoffs. This reference covers exact context constraints, per-tier rate quotas, billing cutoffs, plan restrictions, and API parameter rules governing production deployments.
Claude Fable 5.2 Limits at a Glance
Claude Fable 5.2 enforces strict capacity boundaries across context length, generation volume, API throughput, and organizational spending as documented in the Anthropic model overview.
Developers deploying the model can review general architecture details in the guide on Claude Fable 5.2 explained before provisioning production infrastructure. The baseline technical specifications and operational ceilings are structured as follows:
| Operational Constraint | Baseline Allowance or Ceiling | Enforcement Mechanism | Primary Documentation Source |
|---|---|---|---|
| Context Window | 1M tokens | Default and maximum context window | Anthropic model documentation |
| Maximum Output Length | 128,000 tokens | Generation limit per API completion turn | Anthropic release notes |
| Start Tier Rate Limits | 4,000 RPM / 500,000 ITPM / 100,000 OTPM | Separate bucket for Claude Fable 5.2 requests | Anthropic rate limits guide |
| Build Tier Rate Limits | 10,000 RPM / 1,500,000 ITPM / 300,000 OTPM | Separate bucket for Claude Fable 5.2 requests | Anthropic rate limits guide |
| Scale Tier Rate Limits | 13,000 RPM / 4,000,000 ITPM / 800,000 OTPM | Separate bucket for Claude Fable 5.2 requests | Anthropic rate limits guide |
| Monthly Spend Ceilings | $500 (Start), $1,000 (Build), $200,000 (Scale) | Hard account pause until 00:00 UTC on the first day of the next month | Anthropic rate limits guide |
| Max and Ultra Plan Quota | Up to 50% of weekly plan allowances | Not published | Anthropic pricing terms |
| Pro Plan Entitlement | Usage credits only (standard access) | Requires separate credit balance after promotion | Anthropic pricing terms |
Context and Output Limits
Claude Fable 5.2 provides a default and maximum context window of 1 million tokens alongside a 128,000-token output limit per individual API call.
The 1M-token context window uses the tokenizer originally introduced with Claude Opus 4.7. Anthropic bills the full 1 million-token context window at standard per-token pricing rather than imposing a premium multiplier on extended prompts, as outlined on the Anthropic pricing page. A 900k-token request costs the same per token as a 9k-token request at the standard $8 per million tokens (MTok) rate.
This linear pricing structure differs from competing large context models. For instance, developer documentation for OpenAI GPT-6 Astra specifies that prompts with more than 272K input tokens are subject to a 2x input rate multiplier and a 1.5x output rate multiplier across the entire request.
Claude Fable 5.2 supports up to 128,000 output tokens per completion.
Claude Fable 5.2 has adaptive thinking always on.
API Rate Limits by Tier
Anthropic segregates Claude Fable 5.2 into a dedicated API rate limit bucket that does not share capacity with Claude Fable 5.1 or Claude Fable 5.
Throughput is managed across three standard tiers: Start, Build, and Scale. Each tier sets explicit ceilings on requests per minute (RPM), input tokens per minute (ITPM), and output tokens per minute (OTPM). Teams requiring throughput beyond standard tiers can compare their baseline needs against Claude Opus 5.5 limits, which Anthropic recommends as the daily driver before moving up to Fable 5.2 for specialized complex workloads.
| Developer Account Tier | Requests per Minute (RPM) | Input Tokens per Minute (ITPM) | Output Tokens per Minute (OTPM) | Monthly Spend Cap |
|---|---|---|---|---|
| Start Tier | 4,000 RPM | 500,000 ITPM | 100,000 OTPM | $500 |
| Build Tier | 10,000 RPM | 1,500,000 ITPM | 300,000 OTPM | $1,000 |
| Scale Tier | 13,000 RPM | 4,000,000 ITPM | 800,000 OTPM | $200,000 |
| Custom Tier | Contact Sales | Contact Sales | Contact Sales | None |
A key technical operational rule is that ITPM metering is cache-aware: Anthropic counts only uncached input tokens against your minute-by-minute rate allowance. Prompt tokens read from Anthropic's prompt cache (at the discounted $0.20 per MTok rate) do not consume ITPM quota.
For organizations using the trusted-access Claude Mythos 5.2 model, rate limiting operates differently. Mythos 5.2, Mythos 5.1, and Mythos 5 share a single combined rate limit bucket.
Spend Caps and What Happens at the Cap
Monthly account spend caps automatically stop API activity once an organization exhausts its designated tier budget.
Anthropic enforces monthly spend caps of $500 for Start tier accounts, $1,000 for Build tier accounts, and $200,000 for Scale tier accounts, while Custom tier accounts operate without automated hard stops.
When an organization reaches its monthly spend cap, usage pauses at the cap. Usage pauses completely across all keys until 00:00 Coordinated Universal Time (UTC) on the first day of the subsequent calendar month.
To avoid production outages, organizations must monitor monthly spend trajectories proactively. Teams anticipating burst workloads or expanding agent fleets must request an account tier upgrade or contact sales well before their month-to-date invoicing reaches the threshold.
Plan Usage Limits
Subscription plan limits govern access to Claude Fable 5.2 across end-user applications, chat workspaces, and team seats.
On permanent commercial terms published by Anthropic, Claude Free tier users receive no access to the Fable model family. Pro tier subscribers paying $20 monthly (or $17 monthly on annual billing) access Fable 5.2 strictly through separately purchased usage credits. Claude Max ($100 per month) and Claude Ultra ($200 per month) accounts include Fable 5.2 usage, but restrict activity to a ceiling of up to 50% of the account's total weekly capacity allowances.
Team subscription plans provide Claude Fable 5.2 at up to 50% of weekly usage limits for premium seat holders. Standard seats do not include native Fable allocations and require usage credits. Enterprise agreements provide fully integrated Fable 5.2 access across self-serve and sales-assisted enterprise contracts.
Prospective users can review availability details in the guide answering is Claude Fable 5.2 free. Anthropic announced a dated launch promotion starting November 9, 2026. Under this promotion, from November 9 to 16, 2026, Pro plans and standard Team seats get all Fable models included up to 50% of weekly usage limits through November 16, 2026 at 11:59 pm Pacific Time (PT), along with one limit reset expiring at that exact cutoff.
Subscribers on Max, Ultra, and premium Team seats with active billing prior to November 9, 2026 at 10:00 am PT get one to three limit resets based on subscription tenure (under 6 months receive one reset, 6 to 12 months receive two, and over 12 months receive three) that expire March 8, 2027 at 11:59 pm PT. A reset restores exhausted weekly allowances but does not alter the underlying model rate limits.
What Else Can Stop a Request
API requests to Claude Fable 5.2 can fail due to specific parameter mismatches, tool version deprecations, privacy flags, or content safety classifications even when account rate limits remain open.
Claude Fable 5.2 runs adaptive thinking on every request. Passing thinking: {"type":"enabled"} with a static token budget or thinking: {"type":"disabled"} results in an immediate HTTP 400 client error. Developers must omit the thinking field entirely or supply {"type":"adaptive"} alongside an allowed effort parameter (low, medium, high, xhigh, or max).
Similarly, forced tool use configurations (tool_choice set to any or bound to a specific required tool) are unsupported on Fable 5.2 and return HTTP 400 errors. Omitting sampling parameters like temperature, top_p, and top_k is also required.
Breaking architectural changes also apply to conversational history and tooling. Under Anthropic's prefix check rules, sending back an existing Fable 5.2 thinking block after altering the system prompt, tool definitions, or preceding messages returns a 400 error unless the client explicitly enables the beta header with prefix_mismatch_behavior set to drop_block. Legacy computer use tools like computer_20251124 return HTTP 400 across all environments except Amazon Web Services (AWS) Bedrock. Developers must migrate to computer_toolset_20260801.
Requests can also halt due to privacy configurations or safety refusals. Claude Fable 5.2 requires 30-day data retention by default as documented on the Anthropic covered models portal. Organizations configured for Zero Data Retention (ZDR) will receive request errors when calling Fable 5.2 unless expressly authorized under a transitional agreement. When safety classifiers trip, the model halts generation and returns an HTTP 200 response with stop_reason: "refusal".
Engineering teams must plan production workloads around the Start tier throughput ceiling of 100,000 output tokens per minute or upgrade to Build or Scale tiers. Treat the $500 Start spend cap as a monthly pause, and prepare consumer plan workflows for the 50% weekly quota limit. Review your API integration parameters and rate tier allocations today to ensure your application workflows operate smoothly within published Claude Fable 5.2 limits.
Frequently Asked Questions
- Claude Fable 5.2 features a 1 million-token context window that supports both text and image inputs. Anthropic bills the full 1M-token context at the standard rate of $8 per million input tokens, without imposing tiered rate penalties on long-context prompts.
- API rate limits for Claude Fable 5.2 depend on developer tiers. Start tier provides 4,000 RPM, 500,000 ITPM, and 100,000 OTPM. Build tier provides 10,000 RPM, 1.5M ITPM, and 300,000 OTPM. Scale tier provides 13,000 RPM, 4M ITPM, and 800,000 OTPM. Input tokens read from prompt cache do not count against your ITPM rate limit.
- Model refusals are processed by safety classifiers and return an HTTP 200 response with
stop_reason: "refusal". Billing for a pre-output refusal depends on its category, while rate limit impact is not published. - Anthropic defines rate limits across Start, Build, and Scale tiers. Teams needing throughput beyond the Scale tier (13,000 RPM and 800,000 OTPM) must contact Anthropic sales to negotiate a custom tier.
- Yes, Claude Mythos 5.2 shares a single combined rate limit bucket with Claude Mythos 5.1 and Claude Mythos 5, whereas Claude Fable 5.2 operates in its own dedicated bucket separate from earlier Fable models. Mythos 5.2 access is restricted to organizations verified through specialized programs like the Cyber Verification Program.
Request an AI Workflow and Limits Audit
Layer3Labs helps technical teams design API architectures around Claude Fable 5.2 limits. Schedule an evaluation to review rate tier planning, caching configurations, and spend guardrails.
Book a Consultation