Meta 
Meta narrows 2026 capex to $135–145bn
Meta raised the bottom of its 2026 capex range, including finance-lease principal, from $130bn to $135bn.
Source: Meta Platforms (SEC Form 8-K, exhibit 99.1) (opens in a new tab)
Esta sección aún no está traducida. El contenido siguiente se muestra en inglés, tal como se publicó. Leer la versión en inglés.
Meta · Text model · GA
Muse Spark 1.3
US$1.25Input / 1M
US$4.25Output / 1M
1.05MContext
–AA index
–Tokens/s
–Default level
Compare with similar models Open the Model Finder
Figures checked 1 Nov 2026. A dash means not published on the sources we checked, so it stays blank rather than estimated.
Which level to choose
Meta names these settings reasoning_effort. max uses more reasoning tokens at the same per-token price, per Meta. dev.meta.ai (opens in a new tab)
| Level | When to use it, per Meta | AA index | Tokens/s |
|---|---|---|---|
| minimal | Shortest reasoning pass dev.meta.ai (opens in a new tab) | – | – |
| low | Light reasoning dev.meta.ai (opens in a new tab) | – | – |
| medium | Moderate depth dev.meta.ai (opens in a new tab) | – | – |
| high | Deep reasoning dev.meta.ai (opens in a new tab) | – | – |
| xhigh | Deeper reasoning dev.meta.ai (opens in a new tab) | – | – |
| max | Extended reasoning beyond xhigh, at the same per-token price dev.meta.ai (opens in a new tab) | – | – |
Our rule of thumb: Simple or speed-critical work: low. Standard work: the lab's default. Complex work: high. Use xhigh or max only when your own tests show a gain, as OpenAI and Anthropic both advise. developers.openai.com (opens in a new tab)platform.claude.com (opens in a new tab)
Charts
US$ per 1M tokens, standard tier, short prompts. Output left blank where not published.
Show the data as a table
| Model | Input US$/1M | Output US$/1M |
|---|---|---|
| Muse Spark 1.3 | 1.25 | 4.25 |
| GPT-5.4 mini | 0.75 | 4.5 |
| Grok 4.20 | 1.25 | 2.5 |
| Grok 4.20 Multi-Agent (beta) | 1.25 | 2.5 |
| Grok 4.3 | 1.25 | 2.5 |
Full sourced profile
| Lab | Meta |
|---|---|
| Type | Text |
| Status | GA dev.meta.ai (opens in a new tab) |
| Released | 2 Sep 2026 models.dev (opens in a new tab) |
| API access | Yes dev.meta.ai (opens in a new tab) |
| API model ID | muse-spark-1.3 dev.meta.ai (opens in a new tab) |
| API rate limits | – |
| Consumer app plans | – |
| Price at each level | Same per-token price at every level. dev.meta.ai (opens in a new tab) |
| Input per 1M tokens | US$1.25 dev.meta.ai (opens in a new tab)Standard mode. Contributor mode US$0.10 if Meta may train on your content |
| Output per 1M tokens | US$4.25 dev.meta.ai (opens in a new tab)Standard mode. Contributor mode US$0.20 |
| Cached input | US$0.15 dev.meta.ai (opens in a new tab) |
| Batch in / out | – |
| Other pricing | – |
| Context window | 1.05M tokens dev.meta.ai (opens in a new tab) |
| Max output | 131K tokens models.dev (opens in a new tab) |
| Input types | text, image, video, audio, pdf dev.meta.ai (opens in a new tab) |
| Output types | text |
| Weights | Closed dev.meta.ai (opens in a new tab)Open-weight sibling: Muse Spark Open |
| Reasoning | Yes dev.meta.ai (opens in a new tab) |
| Tool use | Yes dev.meta.ai (opens in a new tab) |
| Artificial Analysis index | – |
| Output speed | – |
| Time to first token | – |
| Reasoning levels | minimal, low, medium, high, xhigh, max dev.meta.ai (opens in a new tab) |
| Lab benchmark claims | – |
| Limitations |
|
Tools and MCP connectors that fit
Meta does not document calling MCP servers from its own API. You can still connect tools through an agent or framework. Browse AI tools
Similar alternatives
Same type, closest on quality index and blended price. A similarity, not a ranking.
- GPT-5.4 miniOpenAI · US$0.75 / US$4.5 per 1M
- Grok 4.20xAI · US$1.25 / US$2.5 per 1M
- Grok 4.20 Multi-Agent (beta)xAI · US$1.25 / US$2.5 per 1M
- Grok 4.3xAI · US$1.25 / US$2.5 per 1M
All 109 AI models · AI agents · How we check sources
Latest news from Meta
All AI news Meta
Inside Prometheus: Meta's gigawatt cluster heads for 1 million GPUs
Meta's next Vera Rubin racks run fully liquid-cooled at over 170 kW each. Starting or pausing a training job can swing tens of megawatts in seconds, so Meta uses batteries and power capping.
Meta
Meta's Muse Wonder wants to sort out your insurance claim
Muse Wonder and Muse Wonder Full are Meta's most capable models, built for agent work in law, finance and health, with Active Memory and Active Focus.
Source: Meta (opens in a new tab)