05 — evaluation surface
23 models — sample set
model benchmarks
Cost, intelligence, and time across the gateway fleet — sample figures until live evals land.
unlock access06 — cost per intelligence
lower is better — usd / task
cost per intelligence
Weighted average cost (USD) per intelligence-index task, split by token type. Lower is better.
polarity
lower
models
23
unit
usd
- answer
- reasoning
- cache write
- cache hit
- input
- DeepSeek V4 Flash 0731
$0.04
- Laguna S 2.1
$0.06
- DeepSeek V4 Flash
$0.08
- DeepSeek V4.1 Flash
$0.15
- MiniMax M3
$0.30
- Mistral Large 3
$0.39
- GLM 5.3
$0.57
- GLM 5.2
$0.66
- GLM 5
$0.83
- Kimi K2.7 Code
$0.98
- Kimi K2.6
$0.98
- DeepSeek V4 Pro
$1.02
- GLM 5 Turbo
$1.03
- GLM 5.1
$1.14
- Abliterated Model
$1.16
- Qwen 3.8 Max
$1.58
- Abliterated Model Large
$1.94
- Abliterated Model Large v2
$1.94
- Qwen 3.7 Max
$1.97
- Kimi K3
$3.56
- CyberGLM
$3.75
- STEALTH-MODEL-01
$6.94
- CyberKimi
$7.50
07 — intelligence index
higher is better — composite
intelligence index
Composite of reasoning, coding, knowledge, and agent evals. Higher is better.
polarity
higher
models
23
scale
index
- published
- estimate
- STEALTH-MODEL-01
76
- Kimi K3
75
- DeepSeek V4 Pro
73
- Qwen 3.8 Max
71
- CyberKimi
70
- Qwen 3.7 Max
68
- CyberGLM
66
- Abliterated Model Large v2
64
- Abliterated Model Large
61
- Kimi K2.7 Code
60
- Kimi K2.6
57
- GLM 5
55
- Abliterated Model
54
- GLM 5.1
53
- Mistral Large 3
52
- GLM 5.2
50
- GLM 5 Turbo
48
- DeepSeek V4.1 Flash
46
- GLM 5.3
43
- Laguna S 2.1
41
- MiniMax M3
38
- DeepSeek V4 Flash 0731
36
- DeepSeek V4 Flash
34
08 — intelligence vs time
pareto — preferred band
intelligence vs time
Intelligence index against minutes per task. Upper left is preferred — high quality, low wait.
x
min
y
index
models
23
- preferred band
- pareto
- frontier
- fleet
STEALTH-MODEL-01
76
7.2 min
frontier
Kimi K3
75
8 min
fleet
DeepSeek V4 Pro
73
6 min
frontier
Qwen 3.8 Max
71
4.8 min
frontier
CyberKimi
70
9.2 min
fleet
Qwen 3.7 Max
68
5.2 min
fleet
CyberGLM
66
6.4 min
fleet
Abliterated Model Large v2
64
5.4 min
fleet
Abliterated Model Large
61
5.6 min
fleet
Kimi K2.7 Code
60
3 min
frontier
Kimi K2.6
57
3.8 min
fleet
GLM 5
55
3.6 min
fleet
Abliterated Model
54
4.2 min
fleet
GLM 5.1
53
3.2 min
fleet
Mistral Large 3
52
2.4 min
frontier
GLM 5.2
50
2.8 min
fleet
GLM 5 Turbo
48
2 min
frontier
DeepSeek V4.1 Flash
46
1.8 min
frontier
GLM 5.3
43
1.6 min
frontier
Laguna S 2.1
41
0.6 min
frontier
MiniMax M3
38
1.2 min
fleet
DeepSeek V4 Flash 0731
36
0.8 min
fleet
DeepSeek V4 Flash
34
1 min
fleet