fireworks Provider Benchmarks

Comprehensive performance summary covering 6 models.

This provider hub highlights throughput and latency trends across every fireworks model monitored by LLM Benchmarks. Use it to compare hosting tiers, track regressions, and discover the fastest variants in the catalogue.

Visit fireworks Official Website

Provider Snapshot

Models Tracked

6

Median Tokens / Second

64.23

Median Time to First Token (ms)

0.63

Last Updated

Dec 25, 2025

Key Takeaways

  • 6 fireworks models are actively benchmarked with 2093 total measurements across 2081 benchmark runs.

  • llama-3.1-8b leads the fleet with 101.00 tokens/second, while kimi-k2 delivers 27.00 tok/s.

  • Performance varies by 274.1% across the fireworks model lineup, indicating diverse optimization strategies for different use cases.

  • Median time to first token across the fleet is 0.63 ms, showing excellent responsiveness for interactive applications.

  • The fireworks model fleet shows varied performance characteristics (37.4% variation coefficient), reflecting diverse model architectures.

Fastest Models

ProviderModelAvg Toks/SecMinMaxAvg TTF (ms)
fireworksllama-3.1-8b101.0045.90154.000.44
fireworksmixtral-8x22b78.1032.10107.000.34
fireworksllama-3.3-70b68.909.60132.000.54
fireworksllama-3.1-70b68.4021.6096.800.48
fireworksllama-3.1-405b42.0020.3055.900.56
fireworkskimi-k227.006.1644.601.42

All Models

Complete list of all fireworks models tracked in the benchmark system. Click any model name to view detailed performance data.

ProviderModelAvg Toks/SecMinMaxAvg TTF (ms)
fireworksllama-3.1-8b101.0045.90154.000.44
fireworksllama-3.1-70b68.4021.6096.800.48
fireworksllama-3.1-405b42.0020.3055.900.56
fireworksmixtral-8x22b78.1032.10107.000.34
fireworksllama-3.3-70b68.909.60132.000.54
fireworkskimi-k227.006.1644.601.42

Featured Models

Frequently Asked Questions

Based on recent tests, llama-3.1-8b shows the highest average throughput among tracked fireworks models.

This provider summary aggregates 2093 individual prompts measured across 2081 monitoring runs over the past month.