Meta · Released Sep 2, 2026Muse Spark 1.3
Muse Spark 1.3 is a multimodal reasoning model from Meta for long-running agentic, multi-agent, and coding workflows. It improves long-horizon agent collaboration, instruction following, and coding efficiency relative to Muse Spark 1.2.
- Capability
- 156.9ECI · #17 of 148
- Input
- $1.25per 1M tokens
- Output
- $4.25per 1M tokens
- Context
- 1.05M131K max output
What it costs
- Input
- $1.25per million tokens
- Output
- $4.25per million tokens
- Cached input
- $0.15per million tokens
- Blended (3:1)
- $2.00pricier than 70% of priced models
Typical monthly bills
| Monthly usage | Estimated cost |
|---|---|
| Side project2M input + 0.5M output tokens | $4.63 |
| Team assistant25M input + 5M output tokens | $52.50 |
| Production app250M input + 50M output tokens | $525.00 |
Official Meta API price, as listed on models.dev.
Independent benchmarks
#17 of 148 scored models
The shaded band is Epoch AI’s confidence range (154.7–159.6); the tick marks the median scored model.
Scores from Epoch AI, run independently of Meta.
The details
- Lab
- Meta
- Released
- Sep 2, 2026
- Context window
- 1,048,576 tokens
- Max output
- 131,072 tokens
- Inputs
- Text, Images, PDFs, Audio, Video
- Output
- Text
- Reasoning
- Adjustable effort minimal · low · medium · high · xhigh · max
- Tool calling
- Yes
- Structured output
- Yes
- Weights
- Proprietary
- API model ID
muse-spark-1.3on Meta- Availability
- 12 API providerslisted on models.dev
What Meta claims
Published by the lab at launch. Settings vary, so compare these only with care.
| Benchmark | Score | Setting | Source |
|---|---|---|---|
| GDPval-AA v2 | 1754 Elo | max effort | Source |
| JobBench | 64.9 | max effort | Source |
| OSWorld v2.0 | 66.9 | max effort | Source |
| OSWorld v2.0 | 32 binary completion rate | max effort | Source |
| DeepSearchQA | 90.3 F1 | max effort | Source |
| Agentic IF Index | 57.8 | max effort | Source |
| AutomationBench | 49.6 | max effort | Source |
| MRCR v2 | 98.5 mean sequence-match ratio | max effort | Source |
| MRCR v2 | 98.1 mean sequence-match ratio | max effort | Source |
| DeepSWE v1.1 | 75.4 | max effort | Source |
| SWE-Atlas Codebase QnA | 59.4 | max effort | Source |
| Terminal-Bench v2.1 | 88.8 | max effort | Source |
More from Meta
About Muse Spark 1.3
How much does Muse Spark 1.3 cost?
Muse Spark 1.3 costs $1.25 input / $4.25 output per million tokens (official Meta API price). Cached input is $0.15 per million tokens. At a 3:1 input-to-output mix that is $2.00 per million tokens, more expensive than 70% of the 360 priced models we track.
What is the context window of Muse Spark 1.3?
Muse Spark 1.3 accepts up to 1,048,576 tokens per request and can write up to 131,072 tokens in one response.
How good is Muse Spark 1.3?
Epoch AI gives Muse Spark 1.3 a Capabilities Index score of 156.9 (likely range 154.7–159.6), ranking it #17 of 148 models Epoch has scored. Epoch AI benchmark results: FrontierMath Tiers 1–3 74.4%, OTIS Mock AIME 2024–2025 99.2%.
Is Muse Spark 1.3 open source?
No. Muse Spark 1.3 is proprietary; you use it through Meta’s API or partner platforms.
What inputs does Muse Spark 1.3 support?
Muse Spark 1.3 accepts text, images, PDFs, audio and video and replies in text. It is a reasoning model with minimal, low, medium, high, xhigh and max effort settings, supports tool calling and can return structured JSON output.
When was Muse Spark 1.3 released?
Meta released Muse Spark 1.3 on Sep 2, 2026.
What are the best alternatives to Muse Spark 1.3?
The closest current models from other labs on capability, price and release date are Gemini 3.8 Flash (Google, ECI 156.9, $0.75 / $3.75), GLM-5.3 (Z.ai (Zhipu), ECI 155.8, $1.40 / $4.40), Grok 4.6 (xAI, ECI 156.6, $2.00 / $6.00) and Qwen3.8 Max (Alibaba (Qwen), ECI 156.6, $2.00 / $6.00).