Meta's biggest jump on agents
75.4 percent on DeepSWE 1.1 for end-to-end agentic software engineering and 88.8 percent on Terminal-Bench 2.1, per Meta AI Research.
Meta released Muse Spark 1.3 on September 2, 2026, its most powerful model and, by its own account, the one that brings it level with the frontier. It is a multimodal reasoning model for long-running agentic, multi-agent and coding workflows: text, images, video, audio and files in, a context window of 1,048,576 tokens, and about 20 percent fewer tool calls than its predecessor for the same job. On avots.ai you use Muse Spark 1.3 and its budget Contributor tier pay per use, no plan, and the same balance also covers GPT-6 Astra, Claude Fable 5.1, Gemini and 40+ image, video and music models.
| Muse Spark 1.3 | Muse Spark 1.3 Contributor | GPT-6 Astra | |
|---|---|---|---|
| Best for | agentic coding, long multi-agent runs | experiments, learning, cheap agent loops | the hardest reasoning |
| Price per 1M in / out | $1.25 / $4.25 | $0.10 / $0.20 | $10 / $50 |
| Context window | 1,048,576 tokens | 1,048,576 tokens | 1.05M tokens |
| Input | text, images, video, audio, files | text, images, video, audio, files | text, images, PDFs |
| Terminal-Bench 2.1 | 88.8% | same model | — |
| On avots.ai | pay per use | pay per use | pay per use |
Contributor is the same model at a twelfth of the input price, meant for experimentation and early-stage agents. Start there, switch to Standard when a hard task deserves it, and put GPT-6 Astra or Claude Fable 5.1 on the same job to compare. One balance covers all of them.
75.4 percent on DeepSWE 1.1 for end-to-end agentic software engineering and 88.8 percent on Terminal-Bench 2.1, per Meta AI Research.
About 20 percent fewer tool calls and 25 percent fewer tokens than Muse Spark 1.2 to finish the same coding work, which is money at every step of an agent.
Text, images, video, audio and files in one request, with 98.5 percent on long-context retrieval across the million-token window.
Meta has shipped a Muse Spark every few weeks since spring; 1.3 is the one its chief AI officer called level with the top competitors.
Sign up on app.avots.ai with email, Google or Telegram. Takes a minute.
1,000 tokens cost 0.99 EUR. No plan to choose, no renewal, the balance sits there until you use it.
Choose a Muse Spark 1.3 tier in the model picker, or let Agent Avots route the task for you. Every reply shows its token cost.
app.avots.ai on desktop and mobile, no install needed.
Muse Spark 1.3 in @AvotsAIbot, same balance, from any phone.
Use it inside Claude Desktop, Cursor or Cline via the MCP server.
The OpenAI compatible API takes your existing client, one key for every model.
1,000 tokens cost 0.99 EUR and a Contributor reply costs a couple of them. The same balance unlocks Standard, GPT-6 Astra, Claude Fable 5.1, Gemini, Veo video and Nano Banana images.
Sign up and chat ✦Every LLM on avots.ai and which one to pick per task.
OpenAI's new flagship and its Pro mode, pay per use.
Anthropic's Mythos-class model, generally available.
Moonshot's frontier open model, number one at coding, 1M context.
Open-weight MoE, 1M context, top LiveCodeBench coding scores.
What to use instead of ChatGPT Plus and how one balance replaces it.
avots.ai runs 40+ AI models for chat, image, video and audio on one balance. See the full list on the Tools page or open the app and start creating.