vs
0:0
overall capability wins (2 ties across 2 scored). GPT-5 leads overall.
- Pricing3:0
Verdict
GPT-5 is ~8.2x cheaper on a blended token basis than GPT-5.6 Cyber.
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- GPT-5
- $3.75
- GPT-5.6 Cyber
- $31.25
GPT-5 is about 8.3x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
OpenAI
OpenAI
Family
GPT-5.x
GPT-5.x
License
Proprietary
Proprietary
Open weights
No
No
Release
Aug 7, 2025
Aug 10, 2026
Knowledge cutoff
2024-09
2026-02-16
API / provider
OpenAI
OpenAI
Modalities
text, image → text
text, image → text
Specs
Context window
400K
400K
Max output
128K
128K
Parameters
—
—
Pricing
Input $/1M
$1.25
$12.50
Output $/1M
$10
$75
Blended $/1M (3∶1)
$3.44
$28.13
Speed
tok/s
90
-
TTFT (s)
0.4
-
Reasoning
MMLU-Pro
—
—
GPQA Diamond
85.4
—
Humanity's Last Exam
28.5
—
AIME 2025
94.6
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
40.3
—
AA-LCR v1.1
78.2
—
CritPt
5.7
—
MMMU-Pro
74.2
—
IFBench
73.1
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
74.9
—
SWE-bench Pro
41.8
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
35.2
—
Aider Polyglot
88.0
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
42.9
—
CursorBench
—
—
SWE-Rebench
—
—
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
—
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,435
—
| Metric | GPT-5 | GPT-5.6 Cyber | Delta |
|---|---|---|---|
| Identity | |||
| Organization | OpenAI | OpenAI | — |
| Family | GPT-5.x | GPT-5.x | — |
| License | Proprietary | Proprietary | — |
| Open weights | No | No | — |
| Release | Aug 7, 2025 | Aug 10, 2026 | — |
| Knowledge cutoff | 2024-09 | 2026-02-16 | — |
| API / provider | OpenAI | OpenAI | — |
| Modalities | text, image → text | text, image → text | — |
| Specs | |||
| Context window | 400K | 400K | tie |
| Max output | 128K | 128K | tie |
| Parameters | — | — | — |
| Pricing | |||
| Input $/1M | $1.25 | $12.50 | A +$11.25 |
| Output $/1M | $10 | $75 | A +$65 |
| Blended $/1M (3∶1) | $3.44 | $28.13 | A +$24.69 |
| Speed | |||
| tok/s | 90 | - | — |
| TTFT (s) | 0.4 | - | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 85.4 | — | — |
| Humanity's Last Exam | 28.5 | — | — |
| AIME 2025 | 94.6 | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 40.3 | — | — |
| AA-LCR v1.1 | 78.2 | — | — |
| CritPt | 5.7 | — | — |
| MMMU-Pro | 74.2 | — | — |
| IFBench | 73.1 | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | 74.9 | — | — |
| SWE-bench Pro | 41.8 | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | 35.2 | — | — |
| Aider Polyglot | 88.0 | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 42.9 | — | — |
| CursorBench | — | — | — |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | — | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,435 | — | — |
GPT-5
- Context
- 400K / 128K out
- Parameters
- —
- Price
- $1.25 / $10
- Speed
- 90 tok/s · 0.4s TTFT
- Modalities
- text, image → text
- License
- Proprietary
Original GPT-5 GA release - still a strong price/performance option versus newer premium SKUs.
GPT-5.6 Cyber
- Context
- 400K / 128K out
- Parameters
- —
- Price
- $12.50 / $75
- Speed
- —
- Modalities
- text, image → text
- License
- Proprietary
OpenAI's most capable and most permissive cybersecurity model (Daybreak program) for authorized vulnerability research and exploit validation — a GPT-5.6 Sol derivative at $12.5/$75; gated behind identity verification and legal declarations.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.
These two models have no overlapping published benchmarks in our dataset. Compare specs and pricing instead.