Skip to content
Model infoModelsSnapshot: 15 Oct 2026

Meta / Muse Wonder

Muse Wonder

Multimodal reasoning model for software engineering, long-running agents and computer use. Available in API preview with Standard pricing.

Language models (LLMs)
previewClosed weightsGeneral purposeReasoningCodingAgents
Official documentation (opens in a new tab)Reviewed 13 Oct 2026
10 of 16 specification fields collected
Context window1.05M1,048,576 tokens
Maximum outputNot collectedAwaiting source review
Input / 1M tokens$4.00Meta Model API
Output / 1M tokens$20.00USD · observed 13 Oct 2026

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
1,048,576 tokensSource for Context / input limit (opens in a new tab)
Maximum output
Not collected
Input modalities
Text, Image, Video, Audio, PDFSource for Input modalities (opens in a new tab)
Output modalities
TextSource for Output modalities (opens in a new tab)
Reasoning
SupportedSource for Reasoning (opens in a new tab)
Tool calling
SupportedSource for Tool calling (opens in a new tab)
Structured output
SupportedSource for Structured output (opens in a new tab)
Streaming
SupportedSource for Streaming (opens in a new tab)
Batch processing
Not collected
Fine-tuning
Not collected
Knowledge cutoff
Not collected
Release date
2026-10-12Source for Release date (opens in a new tab)
Parameters
Not collected
Architecture
Not collected
Languages
Multilingual training; exact evaluated language inventory not suppliedSource for Languages (opens in a new tab)
Context caching
SupportedSource for Context caching (opens in a new tab)

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

Meta Model API

Provider source (opens in a new tab)
API model ID
muse-wonder-1.0
Access
Bearer-authenticated API at https://api.meta.ai/v1. Supports Responses, Chat Completions and Messages protocol surfaces.
Regions
Availability is geography-controlled; see Meta geographic-use policy. No data-residency guarantee is inferred.
protocols
POST /v1/responses, POST /v1/chat/completions, POST /v1/messagesSource for protocols (opens in a new tab)
reasoningLevels
minimal, low, medium, high, xhigh, maxSource for reasoningLevels (opens in a new tab)

none is rejected; internal reasoning consumes output budget and is billed as output.

Request quotas
3,000 requests/minute; 4,000,000 tokens/minuteSource for Request quotas (opens in a new tab)

Team-wide per model, not per key. Background starts also have a 600/minute cap.

background
SupportedSource for background (opens in a new tab)

Background job creation cannot also set stream:true. Its body is capped at 1 MiB. This is not evidence for a discounted bulk batch API.

outputBudget
At least 16 output tokens on Responses; input plus requested output must fit contextSource for outputBudget (opens in a new tab)

Maximum generation length is model-configured but not numerically specified in the reviewed documentation.

trainingUse
Not supportedSource for trainingUse (opens in a new tab)

Standard request content is not used to train Meta models.

INPUT / 1M TOKENS$4.00
OUTPUT / 1M TOKENS$20.00

Standard mode. No long-context premium. Reasoning tokens are billed as output; web search is additional. Cached input price: $0.20 / 1M tokens (base tier).

USD · observed 13 Oct 2026 · Pricing evidence (opens in a new tab)
Published billing units · original currency · observed 13 Oct 2026
CategoryUsageRateUnit
ToolsWeb search grounding1$2.501,000 search queries

Rate conditions

  1. Additional to text tokens; fee counts tool search queries, not user questions.
Pricing source (opens in a new tab)

Data & enterprise terms

Standard requests are not used to train Meta models. Terms allow retention for service operation, safety/security, legal and policy purposes without a fixed general duration. Standard mode is not itself a zero-retention guarantee.

Read the service policy (opens in a new tab)

Caching

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

01 / INITIAL INPUTProcess & retain

The system processes eligible input and retains reusable state.

02 / LATER REQUESTCheck available state

Reuse depends on matching input, session state and the cache policy.

03 / CACHE HITReuse & generate

Reusing state can reduce repeated work. Any charges follow the billing rules below.

Meta Model API

muse-wonder-1.0
Documented
Support evidence (opens in a new tab)
How it works
Automatic exact-prefix KV reuseSource for How it works (opens in a new tab)
Minimum reusable input
Not collected

No minimum cacheable token count or cache-block size is published in the reviewed guide.

Lifetime & refresh
In-memory eviction; Responses optionally accepts a 24-hour retention hintSource for Lifetime & refresh (opens in a new tab)

Not guaranteed: entries may be evicted early. The guide gives no fixed inactivity TTL.

How to enable it
No flags required. Optional prompt_cache_key on Responses and Chat Completions groups related requests; prompt_cache_retention is Responses-only.Source for How to enable it (opens in a new tab)
Usage fields
Responses: input_tokens_details.cached_tokens; Chat Completions: prompt_tokens_details.cached_tokensSource for Usage fields (opens in a new tab)

Cached tokens are a subset of input, not extra input. Chained previous_response_id reports new-turn usage rather than all reconstructed history.

Billing rules
Cached input: USD .20 per million tokensSource for Billing rules (opens in a new tab)

The reviewed tariff does not establish a separate cache-write or storage charge.

Limits & data boundaries
Only matching leading tokens are reused. store:false does not disable caching. Hit rate and retention depend on backend state.Source for Limits & data boundaries (opens in a new tab)
Cache reads, writes & storage
ChargeApplies toRate / unit
Cache read1Text tokens · Base tier$0.20 / 1M tokensPricing evidence (opens in a new tab)

Rate conditions

  1. Standard mode. No long-context premium. Reasoning tokens are billed as output; web search is additional.

Observed 13 Oct 2026. Write rates and storage fees are different charges; use the billing rules above.

Caching reviewed 13 Oct 2026

What the measurements say

No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

The record behind the model

Why it is included

Meta identifies this native Muse model in its research or hosted-model catalog. The reviewed release is available through provider-controlled service, subscription or research preview, without a released weight artifact in the reviewed inventory. The September Spark roadmap still describes its weight release as forthcoming. Open-weight Glimmer, SAM and Llama are excluded separately; hosted access alone does not turn those models into closed models.

Classification evidence (opens in a new tab)

Lifecycle dates

Provider release12 Oct 2026Source (opens in a new tab)

Sources reviewed