Muse Spark 1.2 on finance benchmarks
It is third on Finance Agent v2 under the main partial-credit score, first under the strict All-Pass score, and it leads the hardest category, financial modeling.
What it is
Muse Spark 1.2 is a Meta model released August 5, 2026, with a 1M-token context window. Vals AI ran it at its highest reasoning setting (xhigh). Developer documentation listed by Vals: dev.meta.ai.
| Developer | Meta |
|---|---|
| Release date | Aug 5, 2026 |
| Context window | 1M tokens |
| Weights | Private |
| Public API price | $1.25 input / $4.25 output per million tokens, listed (source) |
What published results say
| Benchmark | Score | Rank | Context |
|---|---|---|---|
| Finance Agent v2 | 60.60% | #3 of 78 | Context: n=450 test questions, 3 runs · Oct 12, 2026 leaderboard · Vals AI · ±0.28 SEM |
| Excel Modeling Benchmark | 56.98% | #42 of 74 | Context: n=51 test tasks · Oct 12, 2026 leaderboard · Vals AI · ±3.13 as published, type not stated |
| Tax Agent Bench | 32.80% | #15 of 69 | Context: n=193 test questions · Oct 12, 2026 leaderboard · Vals AI · ±2.75 as published, type not stated |
- Leads the Financial Modeling category of Finance Agent v2 (DCF, LBO, accretion and dilution) at 34.5% Context: n: not disclosed per category · Oct 12, 2026 leaderboard · Vals AI · margin: not disclosed.
- Strict All-Pass score on Finance Agent v2 (every check must pass): 50.88% Context: n=450 test questions, 3 runs · Oct 12, 2026 leaderboard · Vals AI · margin: not disclosed
- Finance Agent v2 cost and time per task: $0.77, 5m04s.
What the benchmarks test: FAB v2, filing-based analyst research questions; EMB, building Excel financial models (LBO, DCF, M&A, comps); Tax Agent, research-grade US corporate tax questions, all-or-nothing grading. All three use held-out test sets and AI judges.
Limits
- Excel modeling is a weak spot: 42nd of 74 on the Excel Modeling Benchmark.
- Its financial modeling category lead on Finance Agent v2 is 34.5% Context: n: not disclosed per category · Oct 12, 2026 leaderboard · Vals AI · margin: not disclosed, which means most modeling answers still miss.
- Benchmarks measure research and modeling answers, not trading results.
Head-to-head
Primary sources
- VerifiedMuse Spark 1.2 model page (Vals AI)Scores, margins, release date, context window and token prices checked October 14, 2026.
- VerifiedFinance Agent v2 leaderboard and methodology (Vals AI)Sample size, grading and update date checked October 14, 2026.
- VerifiedExcel Modeling Benchmark leaderboard and methodology (Vals AI)Sample size, grading and update date checked October 14, 2026.
- VerifiedTax Agent Bench leaderboard and methodology (Vals AI)Sample size, grading and update date checked October 14, 2026.
- LinkedDeveloper documentation (Meta)Linked by Vals AI. Not re-checked by us.
Related stories
- AI Analyst ScoreboardThe best AI analyst scores 65.4% on a Wall Street research test
Last verified: . Prices and scores change; check the primary source before relying on them. We have no affiliate links and receive nothing from the companies listed. Every percentage carries its sample size, period and source (no naked percentages).
Not financial advice. For information and education only. Benchmark scores and vendor descriptions are not trading results and do not predict returns. Nothing here is a recommendation to buy or sell any asset or to subscribe to any product. See our disclaimer.