Meta's first metered model against OpenAI's small tier.
muse-spark · meta
gpt-5.4-mini · openai
Benchmark
| Workload Scenario | Muse Spark | GPT-5.4 mini | Price Delta |
|---|---|---|---|
| 1M input tokens (raw text) | $1.25 | $0.75 | +$0.50 |
| 1M output tokens (generation) | $4.25 | $4.50 | −$0.25 |
| 1M tokens · 70% input / 30% output mix | $2.15 | $1.875 | +$0.275 |
| Standard chat turn (4K in / 800 out, 50% cached) | $0.0084 | $0.00525 | +$0.00315 |
| Monthly scale (10K requests / day) | $2,520.00 | $1,575.00 | +$945.00 |
Negative difference = Muse Spark is cheaper. Positive = GPT-5.4 mini is cheaper.
Scaling Curve
Total cost of a token volume at 70/30 input/output split (uncached). Log-log scale.
Verdict
GPT-5.4 mini has the cheaper input rate, while Muse Spark has the cheaper output rate. The crossover happens when output makes up about 67% of your total tokens.
Below that share (retrieval, summarization, extraction — lots of context in, little text out) GPT-5.4 mini is cheaper. Above it (generation, translation, coding — long completions) Muse Spark wins.
FAQ
On input, GPT-5.4 mini is cheaper ($0.75/M vs $1.25/M). On output, Muse Spark is cheaper ($4.25/M vs $4.50/M). For workloads where more than 67% of tokens are output, the output-cheaper model wins overall.
A chat-style request (4,000 input + 800 output tokens, 50% cached) costs $0.0084 on Muse Spark and $0.00525 on GPT-5.4 mini — GPT-5.4 mini is 1.6× more expensive for that workload.
Muse Spark supports 128,000 tokens (16.4K max output); GPT-5.4 mini supports 1,050,000 (128K max output). GPT-5.4 mini fits 8.2× more context, which matters for long documents and agents.