Updated Sep 6, 4:43 PM GMT+2

Model Analytics

Usage of models across Spettro. See which models are winning real usage, how the mix shifts over time, and where momentum is moving each week.

SuperFast (DeepSeek V4 Flash) leads with 43.1% of token usage, ahead of Kimi K3 at 15.5% and GLM 5.2 at 14.6%.

Total Tokens 34.6B Jun 2026 to Sep 2026

Top Model 24h Kimi K3 485.1M tokens · 50.4%

Top models.

Usage of models across Spettro.

SuperFast (DeepSeek V4 Flash) DeepSeek

14.9B

+7%

Kimi K3 Moonshot

5.4B

+274%

GLM 5.2 Zhipu

+721%

MiniMax M3 MiniMax

3.7B

-2%

SuperSmart (DeepSeek V4 Pro) DeepSeek

3.3B

+11%

Kimi K2.7 Coder Moonshot

1.4B

New

Qwen3.8 Max Qwen

817.4M

+30%

Qwen3.7 Max Qwen

118.9M

New

Qwen3.6 Flash Qwen

39M

New

Qwen3.7 Plus Qwen

5.9M

-94%

GLM 5.1 Zhipu

67.3K

New

Leaderboard.

Full ranking by token usage.

#		Model	Provider	Tokens	Market Share	Δ 7d
01	=	SuperFast (DeepSeek V4 Flash)	DeepSeek	14.9B	43.1%	+7%
02	▲1	Kimi K3	Moonshot	5.4B	15.5%	+274%
03	·	GLM 5.2	Zhipu	5B	14.6%	+721%
04	=	MiniMax M3	MiniMax	3.7B	10.6%	-2%
05	▼1	SuperSmart (DeepSeek V4 Pro)	DeepSeek	3.3B	9.5%	+11%
06	·	Kimi K2.7 Coder	Moonshot	1.4B	3.9%	New
07	=	Qwen3.8 Max	Qwen	817.4M	2.4%	+30%
08	·	Qwen3.7 Max	Qwen	118.9M	0.3%	New
09	·	Qwen3.6 Flash	Qwen	39M	0.1%	New
10	·	Qwen3.7 Plus	Qwen	5.9M	0%	-94%
11	·	GLM 5.1	Zhipu	67.3K	0%	New

Session cost.

Average cost per session.

COST / SESSIONTOKENS / SESSION

$0.001869SuperFast (DeepSeek V4 Flash)

$0.004296Qwen3.6 Flash

$0.008843GLM 5.1

$0.0158Qwen3.8 Max

$0.0160MiniMax M3

$0.0219SuperSmart (DeepSeek V4 Pro)

$0.0236Kimi K2.7 Coder

$0.0255Qwen3.7 Plus

$0.0348GLM 5.2

$0.0370Kimi K3

$0.0463Qwen3.7 Max

Token cost.

Fixed pricing per 1M tokens (input / output).

COST / 1M TOKENS (IN → OUT)

$0.130 / $0.250SuperFast (DeepSeek V4 Flash)

$0.400 / $1.000SuperSmart (DeepSeek V4 Pro)

$0.600 / $2.400Minimax M3

$0.800 / $3.200Qwen3.7 Plus

$0.894 / $3.713Kimi K2.7 Code

$0.660 / $3.961Qwen3.6 Flash

$1.100 / $3.851GLM 5.2

$1.720 / $5.050Qwen3.8 Max

$3.000 / $9.000Kimi K3

Cache hits.

Share of prompt tokens served from cache.

97% of prompt tokens were
served from cache

95.8% median model hit rate

7 models reporting

Cached share of prompt tokens

97.9%SuperFast (DeepSeek V4 Flash)

97.7%SuperSmart (DeepSeek V4 Pro)

96.7%GLM 5.2

95.8%Kimi K3

95.8%Qwen3.8 Max

94.6%MiniMax M3

89.5%Kimi K2.7 Coder

[*] dashed marker = fleet average · only requests with cached tokens are counted; models without cache hits are hidden.