API host · catalogue checked 27 Oct 2026

Together: models, prices and uptime

Together serves 11 of the 78 models we benchmark through OpenRouter, 1 of them on no other host. Prices are OpenRouter's list prices per million tokens. We did not measure speed per host.

Models served11of 78 we test

Cheapest host for2of 10 shared models

Median premium+208%over the cheapest host

Uptime100%last 30 min, average

Most accurate models on Together

Verification score of each model this host serves

1Muse Spark OpenMeta77.5%
2DeepSeek V4.1 Flash 1017DeepSeek73.8%
3DeepSeek V4 Pro 0813DeepSeek71.2%
4GLM 5.3 FlashZ.ai67.5%
5GLM 5.3Z.ai67.5%
6InklingThinking Machines66.2%
7MiniMax M3.1 FlashMiniMax66.2%
8gpt-oss-120bOpenAI58.8%
9Qwen3.8 2.4T A95BQwen57.5%
10MiniMax M3MiniMax57.5%
11Nemotron 3.5 Super VLNVIDIA40.0%

Models on Together

Premium: this host's blended price (3 input : 1 output) against the cheapest host for the same model

Model	Score	Input / 1M	Output / 1M	vs cheapest host	Hosts	Context	Max output	Quantization	Uptime
Muse Spark Open	77.5%	$1.25	$4.25	Cheapest	6	1M	944k	—	100%
DeepSeek V4.1 Flash 1017	73.8%	$0.3	$1.2	+364%	30	1M	944k	—	100%
DeepSeek V4 Pro 0813	71.2%	$1.32	$3.96	+843%	23	1M	944k	—	100%
GLM 5.3 Flash	67.5%	$0.15	$0.5	+406%	34	1M	944k	—	100%
GLM 5.3	67.5%	$1.4	$4.4	+434%	37	1M	944k	—	100%
Inkling	66.2%	$1	$4.05	Only host	1	524k	472k	—	100%
MiniMax M3.1 Flash	66.2%	$0.15	$0.6	+5%	11	1M	944k	—	100%
gpt-oss-120b	58.8%	$0.15	$0.6	+337%	23	131k	118k	—	100%
Qwen3.8 2.4T A95B	57.5%	$2	$6	Cheapest	5	1M	909k	—	100%
MiniMax M3	57.5%	$0.3	$1.2	+35%	12	524k	472k	—	100%
Nemotron 3.5 Super VL	40.0%	$0.15	$0.5	+79%	9	262k	236k	—	100%

All API hosts →

Together: frequently asked questions

Which AI models does Together serve?: Of the models we benchmark, Together serves 11 through OpenRouter: Muse Spark Open, DeepSeek V4.1 Flash 1017, DeepSeek V4 Pro 0813, GLM 5.3 Flash, GLM 5.3, Inkling, MiniMax M3.1 Flash, gpt-oss-120b, Qwen3.8 2.4T A95B, MiniMax M3, Nemotron 3.5 Super VL.
What is the most accurate model on Together?: Muse Spark Open by Meta, with a verification score of 77.5% (62 of 80 questions correct) in our tests.
Is Together the cheapest host?: Together has the lowest list price for 2 of the 10 models it serves that other hosts also serve (Muse Spark Open, Qwen3.8 2.4T A95B). Its median markup over the cheapest host is +208%.
How reliable is Together?: Averaged over the models it serves, Together reported 100% uptime in the 30 minutes before OpenRouter's catalogue was read for our run.

Model

Score

Input / 1M

Output / 1M

vs cheapest host

Hosts

Context

Max output

Quantization

Uptime

Muse Spark Open

77.5%

$1.25

$4.25

Cheapest

944k

—

100%

DeepSeek V4.1 Flash 1017

73.8%

$0.3

$1.2

+364%

944k

—

100%

DeepSeek V4 Pro 0813

71.2%

$1.32

$3.96

+843%

944k

—

100%

GLM 5.3 Flash

67.5%

$0.15

$0.5

+406%

944k

—

100%

GLM 5.3

67.5%

$1.4

$4.4

+434%

944k

—

100%

Inkling

66.2%

$4.05

Only host

524k

472k

—

100%

MiniMax M3.1 Flash

66.2%

$0.15

$0.6

+5%

944k

—

100%

gpt-oss-120b

58.8%

$0.15

$0.6

+337%

131k

118k

—

100%

Qwen3.8 2.4T A95B

57.5%

Cheapest

909k

—

100%

MiniMax M3

57.5%

$0.3

$1.2

+35%

524k

472k

—

100%

Nemotron 3.5 Super VL

40.0%

$0.15

$0.5

+79%

262k

236k

—

100%

Together: frequently asked questions

Which AI models does Together serve?

Of the models we benchmark, Together serves 11 through OpenRouter: Muse Spark Open, DeepSeek V4.1 Flash 1017, DeepSeek V4 Pro 0813, GLM 5.3 Flash, GLM 5.3, Inkling, MiniMax M3.1 Flash, gpt-oss-120b, Qwen3.8 2.4T A95B, MiniMax M3, Nemotron 3.5 Super VL.

What is the most accurate model on Together?

Muse Spark Open by Meta, with a verification score of 77.5% (62 of 80 questions correct) in our tests.

Is Together the cheapest host?

Together has the lowest list price for 2 of the 10 models it serves that other hosts also serve (Muse Spark Open, Qwen3.8 2.4T A95B). Its median markup over the cheapest host is +208%.

How reliable is Together?

Averaged over the models it serves, Together reported 100% uptime in the 30 minutes before OpenRouter's catalogue was read for our run.