CoreWeave vs Fireworks: LLM provider comparison

At a glance

Metric winners

There is no overall score. Each winner answers one specific question using only directly comparable data.

Fastest on shared models

CCoreWeave

1.64× typical advantage

7 exact models with complete recent speed data

Lowest token cost

CCoreWeave

$1.6129 / 1M

7 exact models using a 1K-input/500-output mix

Most models available

CCoreWeave

20 models

Complete catalog coverage across all indexed modalities

Shared-model benchmark summary

Metric	CCoreWeave☆☆☆☆☆(0)	FFireworks☆☆☆☆☆(0)
Speed	CoreWeave7.35 s	Fireworks8.66 s
TTFT	CoreWeave0.60 s	Fireworks1.23 s
TPS	CoreWeave75.0 tok/s	Fireworks67.0 tok/s
Uptime	CoreWeave99.90%	Fireworks98.85%
Blended	CoreWeave$1.6129 / 1M	Fireworks$1.6267 / 1M

Model	CCoreWeave☆☆☆☆☆(0)	FFireworks☆☆☆☆☆(0)
ZA GLM 5.2Winner · Fireworks	CoreWeave Speed— Worst comparable value 20.46 s TTFT— Worst comparable value 1.94 s TPS— Worst comparable value 27.0 tok/s Uptime— Worst comparable value 99.96% Context 262.1K Route coreweave/fp4 Blended— Best comparable value $2.3933 / 1M	Fireworks Speed— Best comparable value 12.10 s TTFT— Best comparable value 1.23 s TPS— Best comparable value 46.0 tok/s Uptime— Best comparable value 99.96% Context 1M Route fireworks Blended— Worst comparable value $2.4 / 1M
M Kimi K2.7 CodeWinner · CoreWeave	CoreWeave Speed— Best comparable value 7.35 s TTFT— Best comparable value 0.68 s TPS— Best comparable value 75.0 tok/s Uptime— Best comparable value 100.00% Context 262.1K Route coreweave/int4 Blended— Best comparable value $1.96 / 1M	Fireworks Speed— Worst comparable value 12.44 s TTFT— Worst comparable value 3.51 s TPS— Worst comparable value 56.0 tok/s Uptime— Worst comparable value 97.49% Context 262.1K Route fireworks Blended— Worst comparable value $1.9667 / 1M
D DeepSeek V4 ProWinner · Fireworks	CoreWeave Speed— Worst comparable value 64.00 s TTFT— Best comparable value 1.50 s TPS— Worst comparable value 8.0 tok/s Uptime— Best comparable value 98.59% Context 1M Route coreweave/fp8 Blended $2.32 / 1M	Fireworks Speed— Best comparable value 11.53 s TTFT— Worst comparable value 1.52 s TPS— Best comparable value 50.0 tok/s Uptime— Worst comparable value 93.96% Context 1M Route fireworks Blended $2.32 / 1M
D DeepSeek V4 FlashWinner · Fireworks	CoreWeave Speed— Worst comparable value 17.27 s TTFT— Best comparable value 0.60 s TPS— Worst comparable value 30.0 tok/s Uptime— Best comparable value 99.85% Context 1M Route coreweave/fp8 Blended $0.1867 / 1M	Fireworks Speed— Best comparable value 8.66 s TTFT— Worst comparable value 1.41 s TPS— Best comparable value 69.0 tok/s Uptime— Worst comparable value 97.91% Context 1M Route fireworks Blended $0.1867 / 1M
M Kimi K2.6Winner · CoreWeave	CoreWeave Speed— Best comparable value 3.30 s TTFT— Best comparable value 0.34 s TPS— Best comparable value 169.0 tok/s Uptime 99.89% Context 262.1K Route coreweave/fp4 Blended $1.9667 / 1M	Fireworks Speed— Worst comparable value 6.91 s TTFT— Worst comparable value 0.42 s TPS— Worst comparable value 77.0 tok/s Uptime N/a Context 262.1K Route fireworks Blended $1.9667 / 1M
ZA GLM 5.1Winner · CoreWeave	CoreWeave Speed— Best comparable value 4.38 s TTFT— Best comparable value 0.50 s TPS— Best comparable value 129.0 tok/s Uptime 100.00% Context 202.8K Route coreweave/fp8 Blended $2.4 / 1M	Fireworks Speed— Worst comparable value 7.19 s TTFT— Worst comparable value 1.02 s TPS— Worst comparable value 81.0 tok/s Uptime 100.00% Context 202.8K Route fireworks Blended $2.4 / 1M
O gpt-oss-20bWinner · CoreWeave	CoreWeave Speed— Best comparable value 4.18 s TTFT— Best comparable value 0.28 s TPS— Best comparable value 128.0 tok/s Uptime— Worst comparable value 87.61% Context 131.1K Route coreweave/fp4 Blended— Best comparable value $0.0633 / 1M	Fireworks Speed— Worst comparable value 7.92 s TTFT— Worst comparable value 0.46 s TPS— Worst comparable value 67.0 tok/s Uptime— Best comparable value 99.78% Context 131.1K Route fireworks Blended— Worst comparable value $0.1467 / 1M

A per-model winner combines blended price and estimated 500-token response time with equal proportional weight. Ties and rows missing either measurement receive no badge. Comparison data calculated Jul 24, 2026, 6:19 PM UTC. Values use one deterministic route per provider and model; missing measurements remain visible as N/a.