Meta prices Muse Ember at $0.10 input and $0.50 output per 1M tokens
Meta has released Muse Ember, the smallest and cheapest model in its Muse family, and priced it on the Meta Model API at $0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens for prompts up to 200,000 tokens. Longer prompts move to a second tier of $0.30 input, $0.012 cached input and $1.50 output per million tokens. The model was announced on November 9, 2026, in a post on Meta's AI research blog, and it is available from launch on Muse Code and on the Meta Model API under the identifier muse-ember-1.0.
Meta describes Ember as a model for repetitive, high-volume agent work: tool calls, coding subtasks, request routing, data extraction and the parsing of long documents. It accepts text, images, audio and video, can operate a computer from screenshots, and offers a 1 million token context window with up to 512,000 output tokens per response. It supports the Responses API, and Meta lists Muse Ember, Muse Spark and Muse Wonder as available on Google Cloud and Oracle Cloud.
The headline input and output prices are identical to those of the two models Meta chose as comparisons, Anthropic's Claude Haiku 5.5 and OpenAI's GPT-6 Luna, which both list $0.10 input and $0.50 output per million tokens on their official pricing pages. Ember separates itself on cached input, which costs $0.004 per million tokens against $0.01 for both rivals, on a long-context threshold that sits at 200,000 tokens, and on a maximum output four times larger than the 128,000 tokens Anthropic and OpenAI allow.
··········
MUSE EMBER API PRICING ON THE META MODEL API.
The prices below come from Meta's launch post. On November 10, 2026, Meta's official pricing and rate-limits page for the Model API did not yet list Muse Ember, so the launch announcement is the only official price source for now. Meta has not published batch prices, rate limits or a Contributor (training-data) variant for Ember.
........
Muse Ember item | Prompts up to 200K tokens | Prompts over 200K tokens |
Input | $0.10 | $0.30 |
Cached input | $0.004 | $0.012 |
Output | $0.50 | $1.50 |
Batch | Not published | Not published |
Context window | 1M tokens | 1M tokens |
Max output | 512K tokens | 512K tokens |
........
The long-context tier triples every rate once a prompt passes 200,000 tokens. That makes Ember different from the rest of Meta's paid lineup: the pricing page states that Muse Spark and Muse Wonder charge the same rate whatever the context size.
··········
MUSE EMBER VS MUSE SPARK 1.3 AND MUSE WONDER PRICES.
Ember has no direct predecessor, so the closest "before" is the cheapest standard model Meta sold until November 9. On the official pricing page, Muse Spark 1.3 costs $1.25 per million input tokens, $0.15 cached and $4.25 output in Standard mode. Muse Wonder, Meta's top model, costs $4.00, $0.20 and $20.00. Ember therefore cuts the entry price of a standard Meta model by 92% on input and about 88% on output.
........
Meta model (Standard mode) | Input | Cached input | Output | Long-context surcharge |
Muse Ember | $0.10 | $0.004 | $0.50 | 3x above 200K tokens |
Muse Spark 1.3 | $1.25 | $0.15 | $4.25 | None |
Muse Spark 1.3 Contributor | $0.10 | $0.002 | $0.20 | None |
Muse Wonder | $4.00 | $0.20 | $20.00 | None |
........
The Contributor variant of Muse Spark 1.3 remains cheaper than Ember on output, at $0.20 per million tokens. That discount comes with a condition: Contributor mode gives Meta permission to use the customer's content to train future models, and its rate limit is 150 requests per minute against 3,000 for the standard Spark models. Ember is sold on standard terms, which makes it the lowest-cost Meta option for companies that cannot hand their prompts over for training. Our earlier coverage of Muse Spark 1.3 and Muse Code describes the model Ember now undercuts.
··········
MUSE EMBER VS CLAUDE HAIKU 5.5 VS GPT-6 LUNA API PRICING.
All prices in the next table were checked on the official Anthropic and OpenAI pricing pages and model pages on November 10, 2026, in USD per million tokens. Claude Haiku 5.5 was released on October 7, 2026; GPT-6 Luna is the small model OpenAI also uses in its Decisions API.
........
Item | Muse Ember | Claude Haiku 5.5 | GPT-6 Luna |
Input | $0.10 | $0.10 | $0.10 |
Cached input | $0.004 | $0.01 | $0.01 |
Cache write | Not published | $0.125 (5 min), $0.20 (1 h) | $0.125 |
Output | $0.50 | $0.50 | $0.50 |
Batch input / output | Not published | $0.05 / $0.25 | $0.05 / $0.25 |
Long-context threshold | 200K tokens | 100K tokens | 272K tokens |
Long-context input / output | $0.30 / $1.50 | $0.50 / $2.50 | $0.20 / $0.75 |
Context window | 1M | 1M | 1.05M |
Max output | 512K | 128K | 128K |
........
The comparison is not fully like-for-like because each company uses its own tokenizer, so the same text produces a different token count on each model. Anthropic, for example, says Haiku 5.5 counts roughly 30% more tokens than Haiku 4.5 for identical text. GPT-6 Luna also differs in how it applies its long-context rate: OpenAI doubles input and cache prices above 272,000 tokens and multiplies output by 1.5, while Anthropic applies a fivefold rate to every category above 100,000 tokens.
··········
[[ADPLUS_300x250_1]]
WHAT A HIGH-VOLUME AGENT WORKLOAD COSTS ON MUSE EMBER.
To make the prices concrete, take a routing and extraction pipeline that handles 200,000 requests a day for 30 days, or 6 million requests a month. Each request sends 3,000 input tokens, of which 2,000 are a shared system prompt and tool definitions served from cache, and returns 300 output tokens. Cache-write charges are left out because the shared prefix is written rarely compared with the number of reads, and all prompts stay in the short-context tier.
........
Model | Monthly cost with caching | Monthly cost without caching | Cost per 1,000 requests (cached) |
Muse Ember | $1,548 | $2,700 | $0.258 |
Claude Haiku 5.5 | $1,620 | $2,700 | $0.270 |
GPT-6 Luna | $1,620 | $2,700 | $0.270 |
Muse Spark 1.3 Standard | $16,950 | $30,150 | $2.825 |
Muse Spark 1.3 Contributor | $984 | $2,160 | $0.164 |
........
The arithmetic for Ember is 6 million requests multiplied by (1,000 uncached tokens at $0.10 plus 2,000 cached tokens at $0.004 plus 300 output tokens at $0.50) per million, which gives $1,548. Without caching the three small models cost exactly the same, $2,700 a month, because their list prices are identical. With a two-thirds cache hit share Ember saves $72 a month, about 4.4%, purely through its lower cached-input rate. Against Muse Spark 1.3 in Standard mode, moving this workload to Ember cuts the monthly bill from $16,950 to $1,548.
The picture changes for asynchronous work. OpenAI's batch rates for GPT-6 Luna ($0.05 input, $0.005 cached, $0.25 output) bring the same workload to $810 a month, roughly half of Ember's cost, and Anthropic's Batch API halves Haiku 5.5's input and output prices as well. Until Meta publishes a batch discount, jobs that can wait hours for results remain cheaper on the two rivals.
Long prompts give a third result. A single request with a 400,000-token document and a 20,000-token answer costs about $0.15 on Ember, $0.25 on Claude Haiku 5.5 and $0.095 on GPT-6 Luna, using each provider's long-context rates. Ember is cheaper than Haiku once Haiku crosses its 100,000-token threshold, and more expensive than Luna, whose surcharge is lighter and starts later.
··········
WHO CAN USE MUSE EMBER AND HOW TO ACCESS IT.
Developers reach Ember through the Meta Model API with the model name muse-ember-1.0, or inside Muse Code, Meta's coding product. Meta also lists the Muse models on Google Cloud and Oracle Cloud, but it has not published Ember's prices on those platforms, and buyers should check each cloud's own listing before committing. The Model API page applies rate limits per team rather than per key, at 3,000 requests and 4 million tokens per minute for each standard Muse Spark and Muse Wonder model; Ember's limits were not on the page on November 10.
The launch post does not say whether Ember powers features in the consumer Meta AI app or in Muse, and it does not mention regional restrictions. Our guide to Meta AI API access and developer tools explains how the developer side of Meta's AI products is organized, though it predates the Muse price list.
For teams already running Claude Haiku 5.5 or GPT-6 Luna, a switch does not change the per-token budget for uncached traffic. The gains come from heavy prompt caching, from responses longer than 128,000 tokens, and from long prompts that would otherwise fall into Anthropic's higher tier. Teams relying on batch processing or on published rate limits have less reason to move today.
··········
MUSE EMBER BENCHMARKS REPORTED BY META.
Meta published benchmark results run at maximum reasoning effort, some by Meta and some marked as Artificial Analysis evaluations. These are vendor-reported figures and had not been reproduced independently at the time of writing.
........
Benchmark | Muse Ember | Claude Haiku 5.5 | GPT-6 Luna |
Artificial Analysis Intelligence Index | 36 | 36 | 28 |
Terminal-Bench 4.0 | 40.9% | 39.2% | 16.4% |
Coding Agent Index-AA | 46 | 37 | 31 |
FrontierCode 2.0-AA | 27.2% | 34.9% | 25.2% |
OSWorld 2.1 (partial credit) | 69.5% | 68.9% | 50.5% |
AutomationBench-AA | 37.8% | 13.9% | 20.2% |
Agents' Last Exam | 61.6% | 57.0% | 57.1% |
........
On Meta's numbers Ember ties Haiku 5.5 on the Artificial Analysis Intelligence Index, matches or beats it on agent and computer-use tests, and trails it on FrontierCode 2.0. Meta also puts the cost of running the full Intelligence Index task set at about $0.14 per task on Ember against $2.39 on Muse Spark 1.3, which scores 37 on the same index. Meta says Ember was the first model trained with Muse Wonder Full and that it stayed below the high-risk thresholds of its Superintelligence Scaling Framework for chemical and biological, cybersecurity and loss-of-control risks.
··········
MUSE EMBER PRICING: CONFIRMED FACTS AND OPEN QUESTIONS.
Meta has confirmed the model name, the two-tier price list of $0.10, $0.004 and $0.50 per million tokens up to 200,000 tokens and $0.30, $0.012 and $1.50 above it, the 1 million token context window, the 512,000-token output ceiling and availability on the Meta Model API and Muse Code from November 9. The official comparison shows Ember priced level with Claude Haiku 5.5 and GPT-6 Luna on input and output, cheaper on cached input, and far below Meta's own Muse Spark 1.3.
Several commercial details are still missing: batch pricing, cache-write charges, rate limits, prices on Google Cloud and Oracle Cloud, and whether a Contributor variant will follow. Those gaps decide the real cost for many buyers, since batch discounts alone make GPT-6 Luna about half the price of Ember on asynchronous jobs. The benchmark lead Meta claims over GPT-6 Luna on agent tasks is large, but until independent tests confirm it, Ember's verified advantage is a cheaper cache and a longer output limit at the same list price as its two main rivals.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org
