AI API pricing this week: Claude Fable 5.2 at $8/$40, Gemini 4 Hydrogen at $1.50/$7.50 and Muse Ember at $0.10

The week of November 9 to 13, 2026 brought 12 verified changes to the API price lists of Anthropic, Google and Meta, led by a 20% cut at the top of Anthropic's range. Anthropic launched Claude Fable 5.2 on November 9 at $8 per million input tokens and $40 per million output tokens, 20% below Claude Fable 5.1, and cut every cache and batch rate by the same proportion. Google released Gemini 4 Hydrogen on November 12 at $1.50 input and $7.50 output with flat pricing across a 1M-token window, moved Gemini 4 Carbon to general availability on the paid tier of the Gemini API at an introductory $2/$10, and introduced the Decisions API, whose gemini-decision-1 model bills $0.10 per million input tokens and nothing for output.

Meta set the lowest price of the week with Muse Ember, released on November 9 at $0.10 per million input tokens, $0.004 for cached input and $0.50 for output on prompts up to 200K tokens. OpenAI's pricing page and API changelog show no new model or price change between November 6 and November 13, and the official pricing and release pages of xAI, Mistral, DeepSeek, Alibaba Model Studio, Moonshot, Amazon Bedrock and Azure OpenAI list no price change effective in the same window. All prices below are in USD per 1M tokens, as listed on the providers' official pricing pages and documentation on November 13, 2026.

··········

ALL API PRICE AND TERMS CHANGES THIS WEEK.

The table lists every change verified this week, one row per provider, model and priced item. "New model" means there is no previous price for the same model; the sections below compare those launches with the provider's previous model in the same tier.

........

Provider	Model	Item	Before	After	Effective date
Anthropic	Claude Fable 5.2	Input / output	$10 / $50 (Fable 5.1)	$8 / $40	Nov 9
Anthropic	Claude Fable 5.2	Cache read / 5m write / 1h write	$0.25 / $12.50 / $20	$0.20 / $10 / $16	Nov 9
Anthropic	Claude Fable 5.2	Batch input / output	$5 / $25	$4 / $20	Nov 9
Anthropic	Claude Fable 5.2 and Mythos 5.2	Fast mode input / output	Not listed for 5.1 models	$16 / $80	Nov 9
Anthropic	Claude Mythos 5.2	Input / output (verified organizations)	$10 / $50 (Mythos 5.1)	$8 / $40	Nov 9
Google	Gemini 4 Hydrogen	Input / output	New model	$1.50 / $7.50	Nov 12
Google	Gemini 4 Hydrogen	Cached input (implicit)	New model	$0.075	Nov 12
Google	Gemini 4 Carbon	Paid-tier Gemini API access	Limited access	GA, $2 / $10 to Nov 30, then $4 / $20	Nov 10
Google	gemini-decision-1	Input / output	New model	$0.10 / $0	Nov 11
Google	gemini-embedding-2-preview	Model routing	Separate preview model	Served by gemini-embedding-2 at its rates	Nov 6
Meta	Muse Ember	Input / cached / output, up to 200K	New model	$0.10 / $0.004 / $0.50	Nov 9
Meta	Muse Ember	Input / cached / output, over 200K	New model	$0.30 / $0.012 / $1.50	Nov 9

........

··········

CLAUDE FABLE 5.2 AND MYTHOS 5.2 API PRICING: 20% BELOW THE 5.1 MODELS.

Anthropic's release notes dated November 9 introduce Claude Fable 5.2 (claude-fable-5-2) for demanding reasoning and long-horizon agentic work, and Claude Mythos 5.2 (claude-mythos-5-2) with the same capabilities for organizations verified through Anthropic's verification programs, such as the Cyber Verification Program. Both models list $8 input and $40 output per million tokens, against $10 and $50 for Claude Fable 5.1 and Claude Mythos 5.1, which remain on the price list. The cut carries through to prompt caching, where cache reads fall from $0.25 to $0.20, 5-minute cache writes from $12.50 to $10 and 1-hour cache writes from $20 to $16, and to the Batch API, where the 50% discount now gives $4 input and $20 output.

Fast mode is new for this generation. Anthropic lists it for Claude Fable 5.2 and Claude Mythos 5.2 on the Claude API at $16 input and $80 output per million tokens, double the standard rate, for up to 2.5 times higher output speed, and the pricing page states that fast mode does not work with the Batch API and stacks with prompt caching and data residency multipliers. Both models have a 1M-token context window billed at the standard per-token rate, 128K maximum output tokens and always-on adaptive thinking.

The terms changed as well. Claude Fable 5.2 and Claude Mythos 5.2 require 30-day data retention and are not available under zero data retention unless Anthropic expressly authorizes it, they enforce the preserved-thinking prefix check for every account (requests that edit an earlier part of the conversation and resend a later thinking block return a 400 error), and they accept computer use only through the newer computer_toolset_20260801 toolset outside Amazon Bedrock. Claude Fable 5.2 is available on the Claude API, Amazon Bedrock, Claude Platform on AWS, Claude Models on Google Cloud and Claude in Microsoft Foundry. Our launch report on Claude Fable 5.2 pricing and plans covers the consumer side, and the Claude Fable 5.2 vs Claude Fable 5.1 comparison details the API breaking changes for teams planning a migration.

··········

GEMINI 4 HYDROGEN API PRICING: $1.50 INPUT, $7.50 OUTPUT ACROSS THE FULL 1M WINDOW.

Google's Gemini API changelog records Gemini 4 Hydrogen (gemini-4-hydrogen) as generally available on November 12, on the free tier with rate limits and on the paid tier at $1.50 per million input tokens and $7.50 per million output tokens. Output pricing includes thinking tokens. Google's model documentation states that Hydrogen uses flat standard pricing on the paid tier across the full context window of 1,048,576 tokens, so a request near the top of the window pays the same per-token rate as a short one, and synchronous output reaches 65,536 tokens per request.

Caching works only implicitly. The documentation lists a 95% discount on cached input, which brings it to $0.075 per million tokens, applied automatically to prompts of at least 4,096 tokens; explicit context caching is not supported. The pages verified this week give no batch or flex rate for Hydrogen. Against Gemini 3.8 Flash, which the official pricing page lists at an introductory $0.75 input and $3.75 output through December 31, 2026 and at $1.50 and $7.50 from January 1, 2027, Hydrogen launches at the price its predecessor will reach in January, with a fourth thinking level (xhigh) added to low, medium and high.

Our Gemini 4 Hydrogen launch report covers availability in the Gemini app and Search, and the Gemini 4 Hydrogen vs GPT-6.1 Terra and Claude Sonnet 5.5 cost comparison works through per-task costs for the three models. The full Gemini price list, including the December 1 increases for Argon and Carbon, is in our guide to Gemini API pricing.

··········

GEMINI 4 CARBON GENERAL AVAILABILITY ON THE GEMINI API PAID TIER.

Gemini 4 Carbon (gemini-4-carbon) moved from limited access to general availability on November 10. Google's announcement and the Gemini API changelog list an introductory price of $2 per million input tokens and $10 per million output tokens through November 30, 2026, the same introductory price as Gemini 4 Argon, and Google states that the price rises to $4 input and $20 output after the introductory period. Developers reach Carbon through the paid tier of the Gemini API in Google AI Studio and through Android Studio, and enterprises through Gemini Enterprise.

The announcement does not list cached input or batch rates for Carbon. For comparison, the official Gemini API pricing page lists Gemini 4 Argon at $2 input, $10 output and $0.10 cached input until November 30, then $4, $20 and $0.20 from December 1, with batch and flex at half the standard rate. Teams testing Carbon at the introductory price should budget for the doubled rate from December 1. The general-availability news is covered in our report on Gemini 4 Carbon for API developers and AI Ultra.

··········

GEMINI DECISIONS API: GEMINI-DECISION-1 AT $0.10 INPUT AND NO OUTPUT CHARGE.

Google introduced the Gemini Decisions API on November 11 with the gemini-decision-1 model, which scores a closed set of options in one forward pass and returns probabilities and confidence scores instead of generated text. The official announcement prices it at $0.10 per million input tokens with zero output tokens billed, and states that the shared input of a request is billed once regardless of how many questions are asked, with each question adding only its own instruction and criteria tokens. A request can carry text, images, audio, video and PDFs up to 256K tokens and up to 1,000 questions, in yes/no, multiple-choice (2 to 255 options) and ordered-rubric formats.

The pricing model suits classification-style workloads such as tool selection, model routing, policy checks and context grading, where a generative model would otherwise bill output and thinking tokens for a one-word answer. Google cites end-to-end latency of 150 to 250 milliseconds, which is a vendor figure. The API is available through the Gemini API in Google AI Studio, with a no-cost tier for prototyping, and support in Gemini on Google Cloud is listed as coming soon. Details are in our report on the Gemini Decisions API launch.

··········

MUSE EMBER API PRICING: META'S CHEAPEST LONG-CONTEXT MODEL.

Meta released Muse Ember (muse-ember-1.0) on November 9 through the Meta Model API and Muse Code, and its official post lists two price tiers by prompt length. Up to 200K tokens, Muse Ember costs $0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens; above 200K tokens the three rates triple to $0.30, $0.012 and $1.50. The model has a 1M-token context window with up to 512K output tokens, supports the Responses API and, according to Meta, is also available on Google Cloud and Oracle Cloud. Meta's post does not mention a batch rate.

The cached-input rate of $0.004 is 4% of the input price, a 96% discount that sits between the 95% listed for Gemini 4 Hydrogen, Claude Sonnet 5.5 and GPT-6.1 Terra and the 97.5% listed for Claude Fable 5.2. Our report on Muse Ember API pricing covers the launch, and the guide to Meta AI API access lists the full Muse price range.

··········

GEMINI EMBEDDING 2 PREVIEW DEPRECATION AND ROUTING.

The Gemini API changelog dated November 6 marks gemini-embedding-2-preview as deprecated and states that, starting November 6, requests to the preview model route to gemini-embedding-2 and are billed at Gemini Embedding 2 rates. Google says the schema, pricing and embedding space are unchanged, so existing vectors do not need to be re-indexed and the only required change is the model string in client code. Prices are unchanged, and the change appears in the table because billing and model identity for the preview endpoint moved to the stable model.

··········

WHAT 2-3 TYPICAL WORKLOADS NOW COST ACROSS THE MAIN MODELS.

The cost table applies the official rates verified on November 13 to three workloads. A customer-support chatbot handles 10,000 conversations a day for 30 days, each request carrying 6,000 input tokens, of which 4,000 tokens (system prompt and knowledge base) are cache hits and 2,000 are fresh, and producing 500 output tokens. A coding-agent session runs 50 turns of 60,000 input tokens each, 90% of them served from cache, with 2,000 output tokens per turn. A long-document job summarizes 1,000 documents of 150,000 input tokens each into 3,000 output tokens, without caching.

The calculation uses the same token counts for every model and excludes cache-write charges, which vary by provider and depend on how often a cache expires. Real token counts differ by tokenizer; Anthropic states that its Claude 4.7 and later models produce about 30% more tokens than earlier Claude models for the same text, so the figures compare price lists on equal token volumes and are not a like-for-like measure of what the same text costs. Output counts include reasoning tokens where a model bills them as output. All prompts stay under 200K tokens, so Muse Ember's lower tier and GPT-6.1 Terra's standard rate apply.

........

Model	Support chatbot, per month	Coding agent, per session	1,000 long documents, standard	1,000 long documents, batch
Claude Fable 5.2	$11,040.00	$6.94	$1,320.00	$660.00
Claude Fable 5.1	$13,800.00	$8.67	$1,650.00	$825.00
Claude Sonnet 5.5	$2,820.00	$1.87	$330.00	$165.00
GPT-6.1 Sol	$2,820.00	$1.87	$330.00	$165.00
GPT-6.1 Terra	$1,410.00	$0.94	$165.00	$82.50
Gemini 4 Hydrogen	$2,115.00	$1.40	$247.50	Not listed
Muse Ember	$139.80	$0.09	$16.50	Not listed

........

The arithmetic for one chatbot request on Claude Fable 5.2 is (2,000 × $8 + 4,000 × $0.20 + 500 × $40) divided by one million, or $0.0368 per request, which over 300,000 requests gives $11,040 a month against $13,800 on Fable 5.1. The coding-agent turn on Fable 5.2 costs (6,000 × $8 + 54,000 × $0.20 + 2,000 × $40) divided by one million, or $0.1388 per turn and $6.94 per 50-turn session. Rates used per million tokens (input / cached input / output) are $8 / $0.20 / $40 for Fable 5.2, $10 / $0.25 / $50 for Fable 5.1, $2 / $0.10 / $10 for Claude Sonnet 5.5 and GPT-6.1 Sol, $1 / $0.05 / $5 for GPT-6.1 Terra, $1.50 / $0.075 / $7.50 for Gemini 4 Hydrogen and $0.10 / $0.004 / $0.50 for Muse Ember; batch rates are half the standard input and output rates where the provider lists them.

Fable 5.2's cut saves $2,760 a month on the chatbot workload and $1.73 per agent session compared with Fable 5.1, yet the model stays roughly four times as expensive as Claude Sonnet 5.5 at the same token volumes. Gemini 4 Hydrogen lands 25% below Sonnet 5.5 and GPT-6.1 Sol and 50% above GPT-6.1 Terra on all three standard-rate workloads, because its three rates sit in the same ratio to theirs. Muse Ember costs about a tenth of GPT-6.1 Terra on all three workloads, which puts high-volume chat, routing and summarization at a different order of cost if its quality fits the task. Gemini 4 Carbon is left out of the table because its announcement gives no cached-input rate.

··········

AI API PRICES IN THE WEEK OF NOVEMBER 9: WHAT BUYERS CAN COUNT ON.

The verified changes this week favour buyers at both ends of the market. At the top, Claude Fable 5.2 and Mythos 5.2 lower Anthropic's most capable tier by 20% on every standard, cache and batch rate, and the new fast mode gives latency-sensitive agents a documented price of $16 and $80. In the middle, Gemini 4 Hydrogen sets a flat $1.50 and $7.50 across a 1M-token window with automatic caching at $0.075, and at the bottom Muse Ember and gemini-decision-1 push input prices to $0.10 for long-context generation and for closed-set decisions.

Several prices remain open. Google's Carbon announcement gives no cached-input or batch rates, and no batch or flex rate for Gemini 4 Hydrogen appears in the model documentation, so those two lines of the price list are still missing for buyers planning bulk workloads. The introductory prices of Gemini 4 Argon and Carbon end on November 30 and the rates double on December 1, while Gemini 3.8 Flash doubles on January 1, 2027. Fable 5.2 and Mythos 5.2 also carry a 30-day retention requirement that excludes zero-data-retention deployments without Anthropic's authorization, a contract term that weighs on regulated buyers as much as the price cut. Last week's changes, including GPT-6 Cyber and the end of Opus 4.8 fast mode on November 30, are in our previous weekly API pricing roundup.

··········

FOLLOW US FOR MORE.

·····

DATA STUDIOS

·····

datastudios.org