Meta's Muse Spark 1.3 Nears Fable 5.1 in Community Benchmarks at Fraction of Cost
Meta's new Muse Spark 1.3 has impressed across several third-party tests, nearly matching Fable 5.1 while running faster and at far lower cost; Scale AI CEO Alexandr Wang repeatedly amplified these results with high praise. All findings so far come from community-built benchmarks rather than official data, but the model is drawing wide attention.
Confirmed
- User SPAC89 tested Muse Spark 1.3 Ultra against Fable 5.1 with identical self-improvement prompts (retry if an independent judge scores below 9.5/10): over 2 hours each, Muse Spark ran 20 self-improvement rounds and 60+ agent calls for under $1 total, which Alexandr Wang called astonishing.
- Another comparison forwarded by Alexandr Wang used the same prompt at max reasoning settings: Muse Spark 1.3 finished in one minute, while Fable 5.1 took 70 minutes and $13.
- A developer's self-built Counter Strike benchmark showed Muse Spark 1.3 nearly tied with Fable 5.1 at much higher speed, recreating a Counter Strike game for just $1.75.
- User elshayib (via Scobleizer's repost) said Muse Spark 1.3 may already outperform the model behind his resident Hermes agent, calling it very good, very fast, and ridiculously cheap for frontier-level intelligence.
Unconfirmed
- All results are personal benchmarks and subjective impressions, not seen in authoritative reviews or Meta's official figures; the model's exact positioning and capability ceiling remain to be verified.
Why it matters
- If Muse Spark 1.3 truly approaches top-tier models at minimal cost, it could dramatically lower the barrier for long-horizon agent tasks and self-improvement loops; the Scale AI CEO's repeated reposts have amplified its community impact.
2026-09-03 ~ 2026-09-04 · 5 related posts
- Episode 1: Meta Launches Muse Spark 1.3: Near-Frontier Performance at ~1/8 the Price(2026-09-03, 53 posts)
- Episode 2: Meta Releases Muse Spark 1.3 as Its Strongest Model(2026-09-03, 2 posts)
- Episode 3: Meta's Muse Spark 1.3 Nears Fable 5.1 in Community Benchmarks at Fraction of Cost(2026-09-03, 5 posts)
- Episode 4: Meta Chief AI Officer Praises muse spark 1.3 Sub-Agent Demo(2026-09-03, 2 posts)
- Episode 5: Meta Muse Spark 1.3 Ranks Second on Agentic Benchmark at a Third of the Cost(2026-09-04, 2 posts)
- Episode 6: Meta Launches Muse Spark 1.3 Max with Major Coding and Agent Gains(2026-09-05, 12 posts)
Primary sources
- Meta's muse spark 1.3 nearly matches Fable 5.1 on CS benchmark at a fraction of the cost — alexandr_wang ·
- Alexandr Wang Touts Meta Muse Spark 1.3: 1 Minute vs Fable 5.1's 70 Minutes and $13 — alexandr_wang ·
- Meta Muse Spark ran 60+ agent loops in 2 hours for under $1, wowing Scale CEO — alexandr_wang ·
- [source] Alexandr Wang Touts Meta Muse Spark 1.3: 1 Minute vs Fable 5.1's 70 Minutes and $13 — alexandr_wang · 2026-09-03
- [source] Meta Muse Spark ran 60+ agent loops in 2 hours for under $1, wowing Scale CEO — alexandr_wang · 2026-09-03
- Meta's Muse Spark ran 20 self-improvement loops with 60+ agent runs in 2 hours for under $1 — TinfoilTricorn · 2026-09-03
- Early Users Praise Meta's Muse Spark 1.3: Fast, Frontier-Level and 'Too Cheap' — Scobleizer · 2026-09-03
- [source] Meta's muse spark 1.3 nearly matches Fable 5.1 on CS benchmark at a fraction of the cost — alexandr_wang · 2026-09-04