Muse Spark 1.3
Meta · released
Meta's flagship hosted model: a multimodal reasoning model with a 1M-token context aimed at long-running agents and coding, priced well below other labs' flagships. It ties for the top score on the legal-research benchmark tracked here.
Best for: Legal and research work · Low-cost long-running agents · Fast output
- Intelligence Index
- 48.1
- #7 of 38 · at Max effort
- Price per 1M tokens
- $1.25 in $4.25 out
- Blended $2 at 3:1
- Context window
- 1M
- Output speed
- 209 tokens/s
- $1.6 measured cost per task
Capability profile
One 0–100 score per area, placing this model between the weakest and strongest tracked model across that area's benchmarks.
- Overall79 · #7 of 38
Based on 3 of 3 benchmarks
- Coding68 · #9 of 39
Based on 5 of 5 benchmarks
- Agents & knowledge work83 · #5 of 40
Based on 6 of 6 benchmarks
- Reasoning & science77 · #9 of 39
Based on 2 of 2 benchmarks
- Knowledge & reliability81 · #10 of 39
Based on 2 of 2 benchmarks
- Vision84 · #7 of 31
Based on 2 of 2 benchmarks
Reasoning effort trade-off
The same model at each effort setting. Lower settings give up a few points for a fraction of the cost and wait.
| Setting | Intelligence Index | Cost / task | First answer |
|---|---|---|---|
| Max | 48.1 | $1.6 | 64s |
| xHigh | 45.1 | $1.37 | 47s |
Measured by Artificial Analysis. First answer is the median wait until the first answer token, including thinking time.
Benchmark results
All third-party results on record. Rank is among the tracked models that have been tested on that benchmark.
Overall
#7 of 38· Max effort· Artificial Analysis
- Vals Index58.2%
#7 of 26· Max effort· $3.79 per test· Vals AI
- Arena · Text1494 ±6
#5 of 33· Max effort· 12,343 votes· Arena
Coding
- Terminal-Bench 4.033.3%
#11 of 38· Max effort· Artificial Analysis
- SciCode58.8%
#9 of 38· Max effort· Artificial Analysis
- Arena · WebDev1656 ±8
#8 of 34· Max effort· 7,678 votes· Arena
- Vibe Code Bench85.9%
#9 of 26· Max effort· Vals AI
- Code Migration48.3%
#7 of 26· Max effort· Vals AI
Agents & knowledge work
- GDPval-AA59.2
#6 of 39· Max effort· Artificial Analysis
- Arena · Agent+4.0 ±0.6
#8 of 29· Max effort· Arena
- Finance Agent v260.0%
#3 of 26· Max effort· Vals AI
#9 of 26· Max effort· Vals AI
- Legal Research Bench55.3%
#1 of 26· Max effort· Vals AI
- Tax Agent Bench72.4%
#5 of 26· Max effort· Vals AI
Reasoning & science
- Humanity's Last Exam48.7%
#8 of 39· Max effort· Artificial Analysis
- CritPt24.9%
#9 of 39· Max effort· Artificial Analysis
- ARC-AGI-2saturatingNot tested
- ARC-AGI-3Not tested
- GPQA Diamondsaturating93.5%
#5 of 27· Max effort· Artificial Analysis
Knowledge & reliability
#10 of 39· Max effort· Artificial Analysis
- AA-LCR83.0%
#9 of 39· Max effort· Artificial Analysis
Vision
- MMMU-Pro82.0%
#7 of 27· xHigh effort· Artificial Analysis
- Arena · Vision1290 ±10
#3 of 19· Max effort· 3,843 votes· Arena
Reported by Meta
Numbers the lab published at launch, run on its own harness and settings. Useful context, but not independently verified and never used in the rankings on this site.
- DeepSWE v1.1 · Max effort75.4%
- OSWorld 2.0 · Max effort, partial score66.9%
- AutomationBench · Max effort49.6%
Specifications
- Lab
- Meta
- Released
- Sep 2, 2026
- Availability
- Preview
- Context window
- 1,048,576 tokens
- Max output
- —
- Accepts
- Text, Images, Audio, Video, PDFs
- Knowledge cutoff
- Not published
- Weights
- Closed
- Reasoning control
- Effort minimal to max
- API model ID
muse-spark-1.3
Self-serve and pay-as-you-go, but Meta still labels the Meta Model API itself a public preview.
Pricing
- Input / 1M tokens
- $1.25
- Cached input / 1M
- $0.15
- Output / 1M tokens
- $4.25
- Blended (3:1)
- $2
- Measured cost / task
- $1.6
Standard tier. A “contributor” tier costs $0.10 in / $0.20 out in exchange for letting Meta train on your prompts.
Official sources
Closest alternatives
The models nearest to this one on overall score.
23 benchmarks tracked. See methodology and sources.