fireworks LLM Benchmarks – Performance & Latency

Provider Snapshot

Models Tracked

Median Tokens / Second

61.64

Median Time to First Token (ms)

0.59

Last Updated

Oct 19, 2025

8 fireworks models are actively benchmarked with 6869 total measurements across 6497 benchmark runs.
llama-3.1-8b leads the fleet with 118.00 tokens/second, while deepseek-r1 delivers 45.50 tok/s.
Performance varies by 159.3% across the fireworks model lineup, indicating diverse optimization strategies for different use cases.
Median time to first token across the fleet is 0.59 ms, showing excellent responsiveness for interactive applications.
The fireworks model fleet shows varied performance characteristics (37.3% variation coefficient), reflecting diverse model architectures.

Provider	Model	Avg Toks/Sec	Min	Max	Avg TTF (ms)
fireworks	llama-3.1-8b	118.00	22.00	194.00	0.33
fireworks	llama-3.1-70b	69.00	10.90	115.00	0.41
fireworks	mixtral-8x22b	62.50	15.30	97.60	0.34
fireworks	llama-3.3-70b	60.20	4.82	124.00	0.52
fireworks	llama-3.1-405b	47.60	3.03	69.30	0.47
fireworks	deepseek-r1	45.50	5.63	99.80	0.76

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

Provider	Model	Avg Toks/Sec	Min	Max	Avg TTF (ms)
fireworks	llama-3.1-8b	118.00	22.00	194.00	0.33
fireworks	llama-3.1-70b	69.00	10.90	115.00	0.41
fireworks	llama-3.1-405b	47.60	3.03	69.30	0.47
fireworks	mixtral-8x22b	62.50	15.30	97.60	0.34
fireworks	deepseek-v3	45.10	1.38	77.10	1.14
fireworks	deepseek-r1	45.50	5.63	99.80	0.76
fireworks	llama-3.3-70b	60.20	4.82	124.00	0.52
fireworks	kimi-k2	45.20	8.70	92.70	0.74