Skip to content
ProviderBench home

Same-model provider benchmark

CoreWeave vs MARA: LLM provider comparison

Compare CoreWeave and MARA on 3 exact shared text models. ProviderBench keeps speed, price, and catalog coverage separate so naturally faster model catalogs cannot distort the result.

CoreWeave

20 indexed models

Headquarters
United States
Server regions
United States
Model types
20 text

MARA

4 indexed models

Headquarters
United States
Server regions
United States
Model types
4 text

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

MARA

1.30× typical advantage

3 exact models with complete recent speed data

Lowest token cost

CoreWeave

$0.53 / 1M

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

Most models available

CoreWeave

20 models

Complete catalog coverage across all indexed modalities

Shared-model benchmark summary

MetricCoreWeave(0)MARA(0)
SpeedCoreWeave6.55 sMARA4.55 s
TTFTCoreWeave0.36 sMARA1.55 s
TPSCoreWeave81.0 tok/sMARA167.0 tok/s
UptimeCoreWeave99.84%MARA94.64%
BlendedCoreWeave$0.53 / 1MMARA$0.5989 / 1M
Green value Better comparable resultRed value Worse comparable result

Visual comparison

Price and performance charts

CoreWeaveMARA
500-token response by shared model

Estimated seconds using recent median response-start and output-speed data. Lower is better.

MARA has the lower typical same-model response ratio across 3 measured models.

Blended token price by shared model

USD per 1 million tokens using a 1,000-input/500-output mix. Lower is better.

CoreWeave has the lower typical price ratio across 3 priced shared models.

Shared text models

3 exact models · newest first

ModelCoreWeave(0)MARA(0)
MiniMax M2.5Winner · MARA
CoreWeave
Speed— Worst comparable value
6.55 s
TTFT— Best comparable value
0.38 s
TPS— Worst comparable value
81.0 tok/s
Uptime— Best comparable value
100.00%
Context
196.6K
Route
coreweave/fp8
Blended— Worst comparable value
$0.6 / 1M
MARA
Speed— Best comparable value
4.01 s
TTFT— Worst comparable value
1.16 s
TPS— Best comparable value
175.0 tok/s
Uptime— Worst comparable value
90.63%
Context
196.6K
Route
mara
Blended— Best comparable value
$0.48 / 1M
DeepSeek V3.1Winner · CoreWeave
CoreWeave
Speed— Best comparable value
9.97 s
TTFT— Best comparable value
0.35 s
TPS— Best comparable value
52.0 tok/s
Uptime
100.00%
Context
161K
Route
coreweave/fp8
Blended— Best comparable value
$0.9167 / 1M
MARA
Speed— Worst comparable value
36.85 s
TTFT— Worst comparable value
5.60 s
TPS— Worst comparable value
16.0 tok/s
Uptime
N/a
Context
131.1K
Route
mara
Blended— Worst comparable value
$0.9667 / 1M
gpt-oss-120bWinner · CoreWeave
CoreWeave
Speed— Worst comparable value
5.92 s
TTFT— Best comparable value
0.36 s
TPS— Worst comparable value
90.0 tok/s
Uptime— Best comparable value
99.68%
Context
131.1K
Route
coreweave/fp4
Blended— Best comparable value
$0.0733 / 1M
MARA
Speed— Best comparable value
4.55 s
TTFT— Worst comparable value
1.55 s
TPS— Best comparable value
167.0 tok/s
Uptime— Worst comparable value
98.66%
Context
131.1K
Route
mara
Blended— Worst comparable value
$0.35 / 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 . Values use one deterministic route per provider and model; missing measurements remain visible as N/a.

CoreWeave vs MARA analysis

How CoreWeave and MARA compare for AI inference

CoreWeave and MARA share 3 indexed text models, including MiniMax M2.5, DeepSeek V3.1, gpt-oss-120b. MARA has the stronger typical response-time result on the directly measured set.

Same-model speed evidence

3 shared models currently have complete response-start and output-speed measurements on both providers. The speed comparison uses per-model ratios before taking the median, so naturally faster model catalogs do not improve the result.

Token pricing on one workload

3 shared models have complete input and output prices on both providers. Prices use the same 1,000-input/500-output-token mix and are normalized to one million tokens for readability.

Catalog and deployment differences

CoreWeave has 20 indexed models and lists 1 published server region; MARA has 4 models and lists 1 region. Verify data residency, privacy terms, limits, and production latency directly before choosing.

Related same-model benchmarks

Compare with other providers

Explore qualified alternatives with the most shared measured models. Recommendations include comparisons for both CoreWeave and MARA.

CoreWeave vs DeepInfra

16 shared model benchmarks

MARA vs NovitaAI

4 shared model benchmarks

CoreWeave vs SiliconFlow

15 shared model benchmarks

MARA vs AtlasCloud

3 shared model benchmarks

CoreWeave vs Parasail

13 shared model benchmarks

MARA vs SambaNova

3 shared model benchmarks