System Card
← leaderboard
OpenAI
GPT Luna
#23
of 28 · general
GPT-6 Luna
1.1M
context
$0.10
/
$0.50
per 1M in / out
2026-09-22
released
2026-05-18
knowledge cutoff
50
100
Coding 28 · #20 of 28
Agents 18 · #25 of 29
Cyber 45 · #1 of 24
Engineering 15 · #21 of 26
Math 42 · #12 of 26
Health 20 · #20 of 24
Legal 19 · #24 of 27
Professional 18 · #24 of 28
Writing 3 · #20 of 21 · one source
Instruct 6 · #24 of 25
Hallucination 40 · #12 of 26
Multimodal 35 · #13 of 24
Coding
28
#20
Agents
18
#25
Cyber
45
#1
Engineering
15
#21
Math
42
#12
Health
20
#20
Legal
19
#24
Professional
18
#24
Writing
3
#20
Instruct
6
#24
Hallucination
40
#12
Multimodal
35
#13
Step
MiMo Flash
GLM Flash
MiniMax Flash
GPT Astra
GPT Sol
History
ECI
AIME
GPQA
APEX-Agents
CritPt
SciCode
WebDev Arena
LiveBench
ECI
156
157
158
159
160
161
Jul 2026
Sep 2026
GPT-5.6 Luna · $0.2/M in · 2026-07-09 · still served · Epoch Capabilities Index 156.5 ECI points
GPT-5.6 Terra · $2/M in · 2026-07-09 · still served · Epoch Capabilities Index 160.1 ECI points
GPT-6 Luna · $0.1/M in · 2026-09-22 · still served · Epoch Capabilities Index 156.8 ECI points
5.6 Luna
5.6 Terra
OpenAI
older → newer
$ / M input tokens
i
3 of 176 priced models · price is a third-party claim, not a vendor quote
Epoch Capabilities Index against input price
Model
$ / M input
ECI points
Released
GPT-5.6 Luna
0.2
156.5
2026-07-09
GPT-5.6 Terra
2
160.1
2026-07-09
GPT-6 Luna
0.1
156.8
2026-09-22
Usage
momentum
usage
users
versions
1.2T
tokens
Sep 23 – Oct 6
▼ 8%
7d
300B
600B
900B
1.2T
Sep 23
Sep 26
Sep 29
Oct 2
Oct 5
Oct 6
Geography
United States
14.0%
Brazil
6.5%
Germany
5.9%
Spain
5.5%
Japan
4.6%
India
3.4%
France
3.3%
Singapore
2.9%
Colombia
2.8%
Argentina
2.7%
rest
91
48.4%
Effort
All independent
$ / task
GPT-6 Luna
GPT-6 Astra
GPT-6.1 Sol
GPT-6 Sol
GPT-5.6 Sol
GPT-5.6 Terra
GPT-5.6 Luna
GPT-5.5
% of best score
0
25
50
75
100
0.1×
0.2×
0.5×
1×
× the best effort’s cost · log scale
None
Low
Med
High
Xhigh
Max
Measured
≈ imputed · amber dot = lab-self-reported among the sources
GPQA Diamond
90.5%
#16
HLE
38.5%
#21
LiveBench
72.0%
#17
SciCode
54.6%
#17
CritPt
19.4%
#13
Terminal-Bench
73.0%
#9
APEX Agents
44.3%
#17
GDPval
46.6%
#20
Arena
1443
#18
AA Index
38.1
#20
Vals Index
51.2
#17
Epoch Capabilities Index
156.8
#11
Neighbors
Stepfun (阶跃星辰)
Step
→ weaker math · pricier
Xiaomi
MiMo Flash
→ weaker multimodal · less factual
Zhipu (智谱)
GLM Flash
→ near-twin
MiniMax
MiniMax Flash
→ stronger agent
DeepSeek Pro
→ follows instructions better · pricier