Muse SparkvsGPT-5.4
Across 8 shared benchmarks, GPT-5.4 leads overall: Muse Spark wins 3, GPT-5.4 wins 5, with 0 ties and an average score difference of -3.74.
Muse Spark
Facebook AI研究实验室 · 2026-04-08 · Reasoning model
GPT-5.4
OpenAI · 2026-03-05 · Multimodal model
Muse Spark3 wins(38%)(63%)5 winsGPT-5.4
Benchmark scores
Grouped by capability, sorted by largest gap within each. 8 shared benchmarks.
General Knowledge
GPT-5.4 2/3| Benchmark | Muse Spark | GPT-5.4 | Diff |
|---|---|---|---|
| ARC-AGI-2 | 42.5028 / 62Thinking (No Tools) | 77.109 / 62Normal (No Tools) | -34.60 |
| HLE | 586 / 172深度思考（无工具、并行） | 52.1021 / 172极高强度思考（工具） | +5.90 |
| GPQA Diamond | 89.5025 / 187Thinking (No Tools) | 92.8011 / 187极高强度思考（无工具） | -3.30 |
AI Agent - Tool Usage
Even 2/2| Benchmark | Muse Spark | GPT-5.4 | Diff |
|---|---|---|---|
| Terminal Bench 2.0 | 5924 / 47Thinking (With Tools) | 75.104 / 47极高强度思考（工具） | -16.10 |
| MCP-Atlas | 82.205 / 27Normal (With Tools) | 70.6014 / 27极高强度思考（工具） | +11.60 |
Math and Reasoning
GPT-5.4 2/2| Benchmark | Muse Spark | GPT-5.4 | Diff |
|---|---|---|---|
| FrontierMath - Tier 4 | 14.6023 / 80Normal (No Tools) | 27.1011 / 80极高强度思考（无工具） | -12.50 |
| FrontierMath | 399 / 60Thinking (No Tools) | 47.605 / 60极高强度思考（无工具） | -8.60 |
Agent Level Benchmark
Muse Spark 1/1| Benchmark | Muse Spark | GPT-5.4 | Diff |
|---|---|---|---|
| τ²-Bench - Telecom | 9220 / 35Thinking (With Tools) | 64.3030 / 35Normal (With Tools) | +27.70 |
Specs
| Field | Muse Spark | GPT-5.4 |
|---|---|---|
| Publisher | Facebook AI研究实验室 | OpenAI |
| Release date | 2026-04-08 | 2026-03-05 |
| Model type | Reasoning model | Multimodal model |
| Architecture | Dense | Dense |
| Parameters | Not available | Not available |
| Context length | 262K | 1M |
| Max output | Not available | 125K |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | Muse Spark | GPT-5.4 |
|---|---|---|
| Text input | Not public | $2.5 / 1M tokens |
| Text output | Not public | $15 / 1M tokens |
| Cache write | Not public | $0.25 / 1M tokens |
One or both models have incomplete public pricing.
Summary
- Muse Sparkleads in:Agent Level Benchmark (1/1)
- GPT-5.4leads in:General Knowledge (2/3), Math and Reasoning (2/2)
- Tied in:AI Agent - Tool Usage
On average across the 8 shared benchmarks, GPT-5.4 scores 3.74 higher.
Largest single-benchmark gap: ARC-AGI-2 — Muse Spark 42.50 vs GPT-5.4 77.10 (-34.60).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.