---
type: "post"
id: "ngrok-ai-gateway-adds-claude-fable-5-2-86827a"
slug: "ngrok-ai-gateway-adds-claude-fable-5-2-86827a"
title: "ngrok AI Gateway adds Claude Fable 5.2 at 20% below Fable 5.1"
description: "ngrok’s AI Gateway added Anthropic’s Claude Fable 5.2 on November 10, with token prices it lists as 20% below Fable 5.1."
retrieval_nugget: "ngrok’s AI Gateway added Anthropic’s Claude Fable 5.2 on November 10, with token prices it lists as 20% below Fable 5.1. The changelog also spells out migration hazards, request-parameter handling, and the model’s always-on adaptive thinking—details that matter when routing production calls and estimating spend."
published_at: "2026-11-16"
updated_at: "2026-11-16"
record_date: "2026-11-16"
date_kind: "published_at"
topics: ["anthropic","claude"]
entities: ["ngrok","AI Gateway","Anthropic","Claude Fable 5.2","Claude Fable 5.1"]
related_patterns: []
novelty_label: "new_fact"
source_url: "https://ngrok.com/changelog/ai-gateway-claude-fable-5-2"
source_urls: ["https://ngrok.com/changelog/ai-gateway-claude-fable-5-2"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:ngrok-ai-gateway-adds-claude-fable-5-2-86827a"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/ngrok-ai-gateway-adds-claude-fable-5-2-86827a/","markdown":"https://newruntime.com/posts/ngrok-ai-gateway-adds-claude-fable-5-2-86827a.md","json":"https://newruntime.com/posts/ngrok-ai-gateway-adds-claude-fable-5-2-86827a.json"}
markdown_url: "https://newruntime.com/posts/ngrok-ai-gateway-adds-claude-fable-5-2-86827a.md"
---

# ngrok AI Gateway adds Claude Fable 5.2 at 20% below Fable 5.1

## Retrieval answer

ngrok’s AI Gateway added Anthropic’s Claude Fable 5.2 on November 10, with token prices it lists as 20% below Fable 5.1. The changelog also spells out migration hazards, request-parameter handling, and the model’s always-on adaptive thinking—details that matter when routing production calls and estimating spend.

## Summary
ngrok’s AI Gateway added Anthropic’s Claude Fable 5.2 on November 10, with token prices it lists as 20% below Fable 5.1. The changelog also spells out migration hazards, request-parameter handling, and the model’s always-on adaptive thinking—details that matter when routing production calls and estimating spend.

## What happened
ngrok’s changelog says Claude Fable 5.2 is available in AI Gateway as `anthropic/claude-fable-5-2`. The listing is dated November 10, 2026, and the announcement describes Fable 5.2 as Anthropic’s model for demanding reasoning and long agentic work. Customers can pay using ngrok credits or attach their own Anthropic API key and be billed by Anthropic. The model accepts requests in Anthropic Messages format and OpenAI chat completions format.

The changelog lists input at $8 per million tokens and output at $40 per million. It compares these with Fable 5.1’s $10 input and $50 output prices, describing each Fable 5.2 price as 20% lower. Cache reads are listed at $0.20 per million tokens. Cache writes cost $10 per million for a five-minute cache and $16 per million for a one-hour cache; the listed Fable 5.1 prices are $0.25, $12.50, and $20 respectively. Fable 5.2 has a one-million-token context window and can return up to 128,000 tokens. It reads text, images, and PDFs, with images billed at the text input rate.

A central operational detail is that adaptive thinking is always on: callers cannot turn it off. Instead, they set effort to low, medium, high, xhigh, or max. ngrok says Anthropic’s default is high, where the model thinks before almost every reply, and thinking tokens are billed as output tokens. The changelog reports that in Anthropic testing, asking the model to answer directly without deliberating did not stop thinking at medium effort or above; for easy requests, it recommends lowering effort. In ngrok’s example, a short question at medium usually produced zero thinking tokens, while a run at high used about 23 before answering.

The update warns that Fable 5.2 returns HTTP 400 for four request types accepted by Fable 5.1, including forced tool use and requests that set thinking to enabled or disabled. The changelog says to omit thinking and set effort instead, and points readers to Anthropic provider documentation for all four cases. AI Gateway passes the errors through; it does not fall back to the next model in a configured list after a 400. Fable 5.2 has no fast mode: the gateway removes `speed: "fast"` and runs the request at standard speed. It also removes `temperature`, `top_p`, and `top_k`, which the model does not accept. The response header `Ngrok-AIG-Removed-Params` names removed fields. For virtual model names, ngrok says placing `anthropic/claude-fable-5-2` first in the list routes the next request to it without a code change, subject to the compatibility caveat.

## What the archive adds
New Runtime’s earlier coverage noted that Fable 5.2 was available through Vercel AI Gateway on November 9 ([Claude Fable 5.2 is live on AI Gateway](https://newruntime.com/posts/claude-fable-5-2-is-live-on-ai-55233b/)). This is a new vendor listing and a new set of ngrok-specific prices and migration details, not the model’s initial launch. An earlier note also covered temporary subscriber usage boosts from Anthropic on November 10 ([Anthropic adds Fable 5.2 launch offers](https://newruntime.com/posts/anthropic-adds-fable-5-2-launch-offers-extra-dcccc1/)); those offers are separate from ngrok’s gateway billing.

## Trend signal
This does not yet fit a pattern New Runtime tracks. A lower per-token price at one gateway is a concrete procurement detail, but it does not establish that agent economics is shifting toward verified task-completion costs or support another tracked pattern on its own.

## Why now
The listing arrives one day after New Runtime’s earlier coverage of Fable 5.2 on Vercel AI Gateway. For teams comparing routing options soon after the model’s launch, ngrok’s published token and cache prices, along with its explicit compatibility warnings, make the gateway choice more concrete.

## Connections to other events
The announcement meets two recent Fable 5.2 developments. New Runtime’s November 9 coverage described availability through Vercel AI Gateway; the ngrok changelog adds another gateway route and its own pricing and request-handling details. New Runtime’s November 10 coverage of Anthropic’s subscriber launch offers concerned temporary usage boosts for eligible Claude plans, rather than payment for API calls through ngrok.

## Why it matters
A 20% lower listed price can change the economics of a workload, but the practical bill still depends on how much the model thinks. Because thinking tokens count as output tokens and adaptive thinking cannot be disabled, teams should include effort settings and reasoning usage in cost comparisons rather than multiplying only visible answer length by the output rate. The changelog’s example makes that distinction measurable: it reports about 23 thinking tokens at high effort for a short request, versus usually none at medium.

Migration behavior matters as much as headline price. A 400 response does not trigger fallback to the next model in an ngrok virtual-model list, so an otherwise resilient routing setup can still fail if an application sends an incompatible request. Teams can treat the switch as a compatibility change: inspect tool-use and thinking parameters, remove unsupported sampling settings, and verify actual routing and billing before making Fable 5.2 the first choice. The response header that identifies stripped parameters gives operators a way to detect silent normalization rather than assume every requested setting took effect.

## Limits and uncertainty
The price table is ngrok’s listing; the announcement does not establish that these prices will remain unchanged or that a given workload will cost less overall. Actual spend depends on input, output, cache use, and thinking-token volume, and no independent workload comparison is reported. The changelog gives examples from ngrok’s runs, not a broad evaluation of model quality or effort-level behavior. It says four request types cause errors and directs readers to Anthropic’s provider documentation for the full set, but the announcement itself names only examples. It also does not quantify latency differences or explain how often parameter stripping affects real applications.

## Skill to build now
Before routing production traffic to Fable 5.2, run a small compatibility check against representative requests from your application. Include forced tool calls, requests with explicit thinking settings, and any temperature or top-p settings; inspect status codes and the `Ngrok-AIG-Removed-Params` response header. Then compare a short, routine request at medium effort with high effort, recording thinking tokens and total cost. Use those results to choose an effort setting for each class of work rather than relying on a prompt asking the model not to deliberate.

## Reader takeaway
ngrok’s Fable 5.2 listing is cheaper per token than its Fable 5.1 listing, but price alone is not the migration decision. Adaptive thinking affects billed output, unsupported parameters can fail without fallback, and some settings are silently removed. Test your real request shapes and measure thinking-token usage before changing a production route.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Anthropic](https://newruntime.com/topics/anthropic.md): Explore the anthropic topic hub.
- topic hub: [Claude](https://newruntime.com/topics/claude.md): Explore the claude topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)
