none reviewed yet · 12 benchmark observations · latest capture Sep 11, 2026
Inspect the evidenceMeta · MODEL PROFILE
Muse Spark 1.3
Teams that want high-taste, very fast, very long-output generation and can accept thin benchmark disclosure
Saved editorial fit for your current task lens; not a benchmark result
How scoring worksVerified rates (Sep 11, 2026) · 100 runs / month at your token settings
Adjust workloadAt a glance
9 facts- Provider
- Meta
- Type
- API model
- Access
- Meta Model API (dev.meta.ai), Muse Code; LLM Stats gateway
- Context
- 1,048,576 tokens
- Maximum output
- 943,700 tokens (LLM Stats)
- License / weights
- Proprietary
- Input modalities
- text, image, video, file, audio
- Output modalities
- text
- Pricing notes
- CONFLICT: Artificial Analysis and MindStudio both report $1.25 in / $4.25 out per 1M; LLM Stats' Meta Model API catalog lists Muse Spark 1.3 at $0.100 in / $0.200 out with $0.002 cached input. Meta's own blog states no price. Scored on the $1.25/$4.25 figure, which two independent sources agree on.
The practical trade-offs
Documented strengths
- Design Arena Website rank 2 (1343) - second-best design-preference score in the set
- AA Intelligence Index 53 at 190.1 tok/s: the best intelligence-per-second combination measured here
- 1M input context with a ~944K max output - by far the largest output budget in this set
- ~20% fewer tool calls and ~25% fewer tokens than Muse Spark 1.2 per Meta
Limitations to check
- No SWE-bench Pro/Verified/LiveCodeBench score for 1.3 - all hard coding evidence is for 1.1/1.2
- Pricing sources conflict by more than 10x ($1.25/$4.25 vs $0.10/$0.20)
- Open weights are promised but not available
- Arena row has only 1,012 votes (+/-21), the widest interval in the top 15
Follow the source
7 source linksEvidence & conditions
7Source links
NoneReviewed records
12Benchmark runs
Sep 11, 2026Latest capture
- Evidence source 1
research.meta.aiDate not recorded - Evidence source 2
artificialanalysis.aiDate not recorded - Evidence source 3
llm-stats.comDate not recorded - Evidence source 4
mindstudio.aiDate not recorded - Evidence source 5
arena.aiDate not recorded - Evidence source 6
benchlm.aiDate not recorded - Evidence source 7
benchlm.aiDate not recorded
Benchmark observations
GDPVal-AA v2 Inspect conditions
- name
- GDPVal-AA v2
- value
- 1754
- unit
- Elo
- source Url
- https://research.meta.ai/articles/introducing-muse-1-3/benchmarks/benchmark-scorecard-v6.webp
- checked At
- 2026-09-08T19:31:52Z
- attribution
- provider-reported
- configuration
- Max effort; 220 tasks; Artificial Analysis Stirrup harness; provider-reproduced score.
JobBench Inspect conditions
- name
- JobBench
- value
- 64.9
- unit
- score
- source Url
- https://research.meta.ai/articles/introducing-muse-1-3/benchmarks/benchmark-scorecard-v6.webp
- checked At
- 2026-09-08T19:31:52Z
- attribution
- provider-reported
- configuration
- Max effort; 65 tasks; mean rubric score; official evaluation with OpenCode.
OSWorld 2.0 (partial) Inspect conditions
- name
- OSWorld 2.0 (partial)
- value
- 66.9
- unit
- score
- source Url
- https://research.meta.ai/articles/introducing-muse-1-3/benchmarks/benchmark-scorecard-v6.webp
- checked At
- 2026-09-08T19:31:52Z
- attribution
- provider-reported
- configuration
- Max effort; version 08.08; 108 workflows; mean partial score on scorecard scale.
OSWorld 2.0 (binary) Inspect conditions
- name
- OSWorld 2.0 (binary)
- value
- 32
- unit
- %
- source Url
- https://research.meta.ai/articles/introducing-muse-1-3/benchmarks/benchmark-scorecard-v6.webp
- checked At
- 2026-09-08T19:31:52Z
- attribution
- provider-reported
- configuration
- Max effort; version 08.08; strict binary workflow completion.
DeepSearchQA Inspect conditions
- name
- DeepSearchQA
- value
- 90.3
- unit
- F1 (0-100)
- source Url
- https://research.meta.ai/articles/introducing-muse-1-3/benchmarks/benchmark-scorecard-v6.webp
- checked At
- 2026-09-08T19:31:52Z
- attribution
- provider-reported
- configuration
- Max effort; 900 questions; common browser/search setup; answer-set F1.
Agentic IF Index (Internal) Inspect conditions
- name
- Agentic IF Index (Internal)
- value
- 57.8
- unit
- index
- source Url
- https://research.meta.ai/articles/introducing-muse-1-3/benchmarks/benchmark-scorecard-v6.webp
- checked At
- 2026-09-08T19:31:52Z
- attribution
- provider-reported
- configuration
- Max effort; Meta internal composite; not a public fixed task set.
AutomationBench Inspect conditions
- name
- AutomationBench
- value
- 49.6
- unit
- %
- source Url
- https://research.meta.ai/articles/introducing-muse-1-3/benchmarks/benchmark-scorecard-v6.webp
- checked At
- 2026-09-08T19:31:52Z
- attribution
- provider-reported
- configuration
- Max effort; public v3; 600 workflows; pass@1.
MRCR v2 (256K-512K) Inspect conditions
- name
- MRCR v2 (256K-512K)
- value
- 98.5
- unit
- score (0-100)
- source Url
- https://research.meta.ai/articles/introducing-muse-1-3/benchmarks/benchmark-scorecard-v6.webp
- checked At
- 2026-09-08T19:31:52Z
- attribution
- provider-reported
- configuration
- Max effort; eight needles; 100 examples; o200k_base bins; sequence-matcher metric.
MRCR v2 (512K-1M) Inspect conditions
- name
- MRCR v2 (512K-1M)
- value
- 98.1
- unit
- score (0-100)
- source Url
- https://research.meta.ai/articles/introducing-muse-1-3/benchmarks/benchmark-scorecard-v6.webp
- checked At
- 2026-09-08T19:31:52Z
- attribution
- provider-reported
- configuration
- Max effort; eight needles; 100 examples; o200k_base bins; sequence-matcher metric.
DeepSWE v1.1 Inspect conditions
- name
- DeepSWE v1.1
- value
- 75.4
- unit
- %
- source Url
- https://research.meta.ai/articles/introducing-muse-1-3/benchmarks/benchmark-scorecard-v6.webp
- checked At
- 2026-09-08T19:31:52Z
- attribution
- provider-reported
- configuration
- Max effort; mini-swe-agent; 113 tasks; functional and regression test pass rate.
SWE-Atlas Codebase QnA Inspect conditions
- name
- SWE-Atlas Codebase QnA
- value
- 59.4
- unit
- %
- source Url
- https://research.meta.ai/articles/introducing-muse-1-3/benchmarks/benchmark-scorecard-v6.webp
- checked At
- 2026-09-08T19:31:52Z
- attribution
- provider-reported
- configuration
- Max effort; mini-swe-agent; 124 public QnA tasks; mean pass@1.
Terminal-Bench 2.1 Inspect conditions
- name
- Terminal-Bench 2.1
- value
- 88.8
- unit
- %
- source Url
- https://research.meta.ai/articles/introducing-muse-1-3/benchmarks/benchmark-scorecard-v6.webp
- checked At
- 2026-09-08T19:31:52Z
- attribution
- provider-reported
- configuration
- Max effort; native coding harness; 89 tasks; isolated sandbox; mean pass@1.
Source reports, measured results, and editorial assessments are separate evidence types. Source dates do not imply a new benchmark run. Capture dates are shown when recorded; otherwise we say the capture date was not recorded. Recorded dates are compared with this catalogue snapshot (Sep 11, 2026), not today’s date.
Method AIS-V2
How this page is assessed
Quality claims follow Method AIS-V2.
Money never buys rank — commercial relationships cannot change rank, eligibility, evidence quality, or publication timing.
Submit or claim a listing is free and does not skip the editorial queue.
Private · this device only