vs
7:2
overall capability wins (2 ties across 11 scored). GPT-6 Astra leads overall.
- Reasoning1:1
- Coding2:1
- Agents4:0
- Composite indices1:0
- Pricing0:3
Verdict
GPT-6 Astra leads coding (widest gap: +14.2 on TermBench 4); GPT-6 Sol is ~5.0x cheaper on a blended token basis than GPT-6 Astra.
- GPT-6 Astra leads coding (widest gap: +14.2 on TermBench 4)
- GPT-6 Sol is ~5.0x cheaper on a blended token basis than GPT-6 Astra
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- GPT-6 Astra
- $22.50
- GPT-6 Sol
- $4.50
GPT-6 Sol is about 5.0x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
OpenAI
OpenAI
Family
GPT-6.x
GPT-6.x
License
Proprietary
Proprietary
Open weights
No
No
Release
Sep 3, 2026
Sep 22, 2026
Knowledge cutoff
2026-04-30
2026-04-20
API / provider
OpenAI
OpenAI
Modalities
text, image → text
text, image → text
Specs
Context window
1.1M
1.1M
Max output
128K
128K
Parameters
—
—
Pricing
Input $/1M
$10
$2
Output $/1M
$50
$10
Blended $/1M (3∶1)
$20
$4
Speed
tok/s
-
-
TTFT (s)
-
-
Reasoning
MMLU-Pro
—
—
GPQA Diamond
96.1
—
Humanity's Last Exam
55.0
48.0
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
57.2
—
AA-Omniscience Accuracy
62.6
—
AA-LCR v1.1
81.0
84.0
CritPt
31.7
—
MMMU-Pro
86.9
—
IFBench
—
—
Chartography
—
53.6
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
88.4
—
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
56.0
58.0
CursorBench
—
—
SWE-Rebench
—
—
NL2Repo-Bench
—
—
DeepSWE
74.1
68.8
WebDev Arena
—
—
Terminal-Bench 4.0
58.2
44.0
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
—
| Metric | GPT-6 Astra | GPT-6 Sol | Delta |
|---|---|---|---|
| Identity | |||
| Organization | OpenAI | OpenAI | — |
| Family | GPT-6.x | GPT-6.x | — |
| License | Proprietary | Proprietary | — |
| Open weights | No | No | — |
| Release | Sep 3, 2026 | Sep 22, 2026 | — |
| Knowledge cutoff | 2026-04-30 | 2026-04-20 | — |
| API / provider | OpenAI | OpenAI | — |
| Modalities | text, image → text | text, image → text | — |
| Specs | |||
| Context window | 1.1M | 1.1M | tie |
| Max output | 128K | 128K | tie |
| Parameters | — | — | — |
| Pricing | |||
| Input $/1M | $10 | $2 | B +$8 |
| Output $/1M | $50 | $10 | B +$40 |
| Blended $/1M (3∶1) | $20 | $4 | B +$16 |
| Speed | |||
| tok/s | - | - | — |
| TTFT (s) | - | - | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 96.1 | — | — |
| Humanity's Last Exam | 55.0 | 48.0 | A +7.0 pts |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | 57.2 | — | — |
| AA-Omniscience Accuracy | 62.6 | — | — |
| AA-LCR v1.1 | 81.0 | 84.0 | B +3.0 pts |
| CritPt | 31.7 | — | — |
| MMMU-Pro | 86.9 | — | — |
| IFBench | — | — | — |
| Chartography | — | 53.6 | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | 88.4 | — | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 56.0 | 58.0 | B +2.0 pts |
| CursorBench | — | — | — |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | 74.1 | 68.8 | A +5.3 pts |
| WebDev Arena | — | — | — |
| Terminal-Bench 4.0 | 58.2 | 44.0 | A +14.2 pts |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | — | — |
GPT-6 Astra
- Context
- 1.1M / 128K out
- Parameters
- —
- Price
- $10 / $50
- Speed
- —
- Modalities
- text, image → text
- License
- Proprietary
OpenAI's September 2026 flagship, the first model OpenAI designates as reaching the Preparedness Framework's Critical cybersecurity capability threshold; rolling out first to Daybreak-program enterprises before wider ChatGPT/API/AWS availability. Standard API pricing is $10/$50 per 1M tokens (input/output), with a Fast mode at $20/$100.
GPT-6 Sol
- Context
- 1.1M / 128K out
- Parameters
- —
- Price
- $2 / $10
- Speed
- —
- Modalities
- text, image → text
- License
- Proprietary
OpenAI's September 2026 general-purpose GPT-6 model for complex coding and agentic workflows. Standard API rates for prompts up to 272K input tokens are $2/$10 per 1M input/output tokens; longer prompts use a higher rate.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.