vs
0:2
overall capability wins (0 ties across 2 scored). o3 leads overall.
- Pricing0:3
- Speed1:1
- Specs0:2
Verdict
o3 offers ~6.1x larger context (200K vs 33K); o3 allows ~12.2x more max output (100K vs 8K); o3 is ~21.4x cheaper on a blended token basis than GPT-4 32K.
- o3 offers ~6.1x larger context (200K vs 33K)
- o3 allows ~12.2x more max output (100K vs 8K)
- o3 is ~21.4x cheaper on a blended token basis than GPT-4 32K
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- GPT-4 32K
- $90
- o3
- $4
o3 is about 22.5x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
OpenAI
OpenAI
Family
GPT-4 / o-series
GPT-4 / o-series
License
Proprietary
Proprietary
Open weights
No
No
Release
Jun 13, 2023
Apr 16, 2025
Knowledge cutoff
-
-
API / provider
OpenAI
OpenAI
Modalities
text → text
text, image → text
Specs
Context window
33K
200K
Max output
8K
100K
Parameters
—
—
Pricing
Input $/1M
$60
$2
Output $/1M
$120
$8
Blended $/1M (3∶1)
$75
$3.50
Speed
tok/s
30
55
TTFT (s)
1
1.2
Reasoning
MMLU-Pro
—
85.3
GPQA Diamond
—
82.7
Humanity's Last Exam
—
20.1
AIME 2025
—
88.9
MATH-500
—
99.2
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
—
38.6
AA-LCR v1.1
—
74.7
CritPt
—
1.1
MMMU-Pro
—
70.1
IFBench
—
71.4
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
58.4
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
80.8
Terminal-Bench 2.1
—
—
Aider Polyglot
—
81.3
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
—
41.0
CursorBench
—
—
SWE-Rebench
—
38.1
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
—
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
1,432
| Metric | GPT-4 32K | o3 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | OpenAI | OpenAI | — |
| Family | GPT-4 / o-series | GPT-4 / o-series | — |
| License | Proprietary | Proprietary | — |
| Open weights | No | No | — |
| Release | Jun 13, 2023 | Apr 16, 2025 | — |
| Knowledge cutoff | - | - | — |
| API / provider | OpenAI | OpenAI | — |
| Modalities | text → text | text, image → text | — |
| Specs | |||
| Context window | 33K | 200K | B +167K |
| Max output | 8K | 100K | B +92K |
| Parameters | — | — | — |
| Pricing | |||
| Input $/1M | $60 | $2 | B +$58 |
| Output $/1M | $120 | $8 | B +$112 |
| Blended $/1M (3∶1) | $75 | $3.50 | B +$71.50 |
| Speed | |||
| tok/s | 30 | 55 | B +25 |
| TTFT (s) | 1 | 1.2 | A +0.20 |
| Reasoning | |||
| MMLU-Pro | — | 85.3 | — |
| GPQA Diamond | — | 82.7 | — |
| Humanity's Last Exam | — | 20.1 | — |
| AIME 2025 | — | 88.9 | — |
| MATH-500 | — | 99.2 | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | — | 38.6 | — |
| AA-LCR v1.1 | — | 74.7 | — |
| CritPt | — | 1.1 | — |
| MMMU-Pro | — | 70.1 | — |
| IFBench | — | 71.4 | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | 58.4 | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | 80.8 | — |
| Terminal-Bench 2.1 | — | — | — |
| Aider Polyglot | — | 81.3 | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | — | 41.0 | — |
| CursorBench | — | — | — |
| SWE-Rebench | — | 38.1 | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | — | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | 1,432 | — |
GPT-4 32K
- Context
- 33K / 8K out
- Parameters
- —
- Price
- $60 / $120
- Speed
- 30 tok/s · 1s TTFT
- Modalities
- text → text
- License
- Proprietary
Long-context GPT-4 SKU (gpt-4-32k-0613) at premium rates — predecessor to 128K Turbo for document workloads.
o3
- Context
- 200K / 100K out
- Parameters
- —
- Price
- $2 / $8
- Speed
- 55 tok/s · 1.2s TTFT
- Modalities
- text, image → text
- License
- Proprietary
OpenAI reasoning model from the o-series; strong on STEM and agentic coding with deliberate chain-of-thought.
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.