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.
Muse Spark 1.3 by meta is offered by 11 providers on this page. Its official list price is $1.25 per 1M input tokens and $4.25 per 1M output tokens.
Artificial Analysis rates it 48.1 on the Intelligence Index, with 75.8 for coding and 55.5 for agentic tasks. Against its lowest blended price of $2.00 per 1M, that is roughly 24 index points per dollar, which is the value ratio the leaderboards rank on. Median output speed is 199 tokens per second, with 58.12s to the first token. Running one task of the Artificial Analysis suite costs about $1.6049, which reflects how many tokens its reasoning consumes rather than the unit price alone.
The context window is 1,048,576 tokens, with an output limit of 131,072 tokens. It supports reasoning, tool use, and structured output. Accepted input modalities are Text, Image, Video, PDF, and Audio.
Quality is independently evaluated by Artificial Analysis. Speed/latency are model-level medians.
| Provider | Tier | Input | Output | Cache read | Cache write | Context | Output limit | Status |
|---|---|---|---|---|---|---|---|---|
| EmpirioLabs AI muse-spark-1-3 | Gateway | $1.25 | $4.25 | $1.00 | — | 1,048,576 | 131,072 | |
| NanoGPT | Gateway | $1.25 | $4.25 | $0.15 | — | 1,048,576 | 943,718 | |
| Kilo Gateway | Gateway | $1.25 | $4.25 | $0.15 | — | 1,048,576 | 943,718 | |
| OpenRouter | Gateway | $1.25 | $4.25 | $0.15 | — | 1,048,576 | 943,718 | |
| MetaOfficial | First-party | $1.25 | $4.25 | $0.15 | — | 1,048,576 | 131,072 | |
| Vercel AI Gateway | Cloud | $1.25 | $4.25 | $0.15 | — | 1,048,576 | 1,048,576 | |
| Opper | Gateway | $1.25 | $4.25 | $0.15 | — | 1,048,576 | 131,072 | |
| OpenCode Zen | Gateway | $1.25 | $4.25 | $0.15 | — | 1,048,576 | 131,072 | |
| DevPass (LLM Gateway) | Gateway | $1.25 | $4.25 | $0.15 | — | 1,048,576 | 131,072 | |
| LLM Gateway | Gateway | $1.25 | $4.25 | $0.15 | — | 1,048,576 | 1,048,576 | |
| AIHubMix | Gateway | $1.38 | $4.68 | $0.17 | — | 1,048,576 | 131,072 |
Sorted by blended price (input×0.75 + output×0.25) asc. The official channel always shows regardless of rank. Whether a gateway's low price is actually usable can't be verified.
Interleaved thinking (reasoning between tool calls) is declared by 1 of 11 providers.
Price history accumulates from each data sync; currently only 1 sample(s) (2026-09-03). Each future sync adds a point, and once accumulated a line is drawn here.
| Name | Conditions | Score | Metric | Source |
|---|---|---|---|---|
| GDPval-AA | harness: Stirrup · variant: max effort · v2 | 1754 | Elo | Source ↗ |
| JobBench | harness: OpenCode · variant: max effort | 64.9 | mean rubric score | Source ↗ |
| OSWorld | harness: Meta internal evaluation framework · variant: max effort · dataset: 08.08 · v2.0 | 66.9 | mean partial score | Source ↗ |
| OSWorld | harness: Meta internal evaluation framework · variant: max effort · dataset: 08.08 · v2.0 | 32 | binary completion rate | Source ↗ |
| DeepSearchQA | harness: Meta browser harness · variant: max effort | 90.3 | F1 | Source ↗ |
| Agentic IF Index | variant: max effort · dataset: Meta internal | 57.8 | score | Source ↗ |
| AutomationBench | variant: max effort · dataset: public v3 task set | 49.6 | pass@1 | Source ↗ |
| MRCR | variant: max effort · dataset: 8-needle, 256K-512K · v2 | 98.5 | mean sequence-match ratio | Source ↗ |
| MRCR | variant: max effort · dataset: 8-needle, 512K-1M · v2 | 98.1 | mean sequence-match ratio | Source ↗ |
| DeepSWE | harness: mini-swe-agent · variant: max effort · v1.1 | 75.4 | pass@1 | Source ↗ |
| SWE-Atlas Codebase QnA | harness: mini-swe-agent · variant: max effort | 59.4 | pass@1 | Source ↗ |
| Terminal-Bench | harness: Muse Code · variant: max effort · v2.1 | 88.8 | pass@1 | Source ↗ |
The same benchmark scores very differently across harness / dataset, so the qualifying conditions must be shown together.
Individual evaluations run by Artificial Analysis, on the same reasoning tier as the intelligence score above. Each benchmark has its own task set and harness, so rows are not comparable with one another. The Intelligence Index above draws on a different, newer set of evaluations.