Meta Model Directory

Muse Spark Open is an 816B total / 42B active open-weight multimodal reasoning model from Meta for coding and long-horizon agentic work, with 1M-token context.

Muse Wonder is Meta's frontier multimodal reasoning model for agents, long-running coding, and computer use, with memory across conversations. The Cursor model id is muse-wonder-1.0, with a 300K default context and a 1M maximum. Pricing is $4/M input, $0.20/M cache read, and $20/M output. There is no cache-write price and no long-context multiplier above 300K. Legacy Cursor plans draw it from the Other Models pool (verified 2026-10-20).

Muse Wonder Full is the most capable model in Meta's Muse Wonder series, built for long-running professional agents and ultra-long-horizon software engineering. It shares Muse Wonder's Active Memory and Active Focus capabilities and generally scores higher on Meta's agentic and coding evaluations.

Muse Spark 1.3 is a multimodal reasoning model released by Meta Superintelligence Labs on 2026-09-02 with 1M context and text/image/video/document input; Meta Model API offers muse-spark-1.3 and muse-spark-1.3-contributor (lower price in exchange for data contribution), and it is already live in the Muse Code coding agent.

Muse Glimmer is a 30-billion-parameter open-weight multimodal model from Meta Superintelligence Labs, distilled from Muse Spark for always-on local agents, tool use, coding, and image understanding.

Muse Spark 1.2 is a multimodal reasoning model released by Meta on 2026-08-05 with 1M context, launched together with the Muse Code terminal coding agent and parallel sub-agents.

Muse Spark 1.1 is a multimodal reasoning model released by Meta on 2026-07-09 with 1M context, improved tool use and computer operation, and the Meta Model API public beta launched alongside it.

Multimodal Llama 4-based safety classifier for moderating prompts and responses

22M-parameter (DeBERTa-xsmall) English classifier that detects prompt injection and jailbreak attempts

86M-parameter (mDeBERTa-base) multilingual classifier that detects prompt injection and jailbreak attempts

Open multimodal Llama for strong reasoning with efficient everyday serving

Open Llama with long-context vision for efficient multimodal agents

Popular open Llama workhorse for multilingual chat, coding, and self-hosting

Compact instruction-tuned Llama model for multilingual dialogue and on-device assistants

Small instruction-tuned Llama model for multilingual dialogue and agentic retrieval

Open multimodal Llama model for image understanding, captioning, and visual QA

Compact open Llama base model for lightweight and on-device use

Small open Llama base model for lightweight text generation and self-hosting

Large open multimodal Llama model for image reasoning, document QA, and captioning

Open Llama instruction model for multilingual chat, reasoning, and coding

Compact open Llama model for lightweight chat, drafting, and self-hosting

Llama 3.1-based safety classifier for moderating prompts and model responses