Muse Spark
Meta's new proprietary flagship — closed-weight, agentic-first
Verdict
Meta shipped this as a new direction, not a Llama update: a closed-weight model served through the Meta Model API, built for long-horizon agentic work, with a 1M-token context window and image, video and document input. Version 1.2 added a coding focus alongside Muse Code, a multi-agent terminal coding tool, and 1.3 is now current, adding a max reasoning setting. The real tradeoff: you lose everything that made Llama useful for self-hosting — no local deployment, no fine-tuning, no community variants. Pick this for Meta's agentic capability specifically; pick Llama if open weights still matter to you.
Other Text Generation & Reasoning
- GPT-6 AstraProduction
OpenAI's most capable model — 1.05M context, 128K output
- GPT-6.1 SolProduction
The OpenAI default — near-Astra quality at a fifth of the price
- GPT-6.1 TerraProduction
OpenAI’s mid tier — half of Sol’s price with the full 1.05M context
- GPT-6 LunaProduction
OpenAI's efficient tier — $0.10/$0.50 with the full 1.05M context
- GPT-5.6Superseded
Previous OpenAI family — Sol/Terra/Luna tiers, superseded by GPT-6
- GPT-5Superseded
OpenAI's 2025 flagship — two families back now