OpenAI · 2026-04-22
o4-pro
192k context · reasoning · multimodal · tools · closed
Capability
90.4
Coding
84.6
Agents
66.2
Science
93.2
TTFT
2.40 s
p50, US-East, streaming
Output speed
48 tok/s
p95 latency
54.0 s
Availability
99.74%
Cost/task
$0.620
Cost/correct task
$0.738
30-day regressions
0
Measured runs
5,410
Last tested 28 min ago
LIVE_MEASUREDEWR 0.033 · human_io 0
TTFT spark · primary endpoint
Category scores
- Math93.8
- Reasoning94.1
- Coding84.6
- Agents66.2
- Multimodal76.4
- Science93.2
- Long context82.8
- Private tasks77.5
Serving stacks
- OpenAI APILIVE_MEASURED2.40 s · 48 tok/s
Reliability on OpenAI API: invalid JSON 1.8% · broken tools 0.7% · replay 96.4%
| Benchmark | Score | $/task | Median s | Evidence |
|---|---|---|---|---|
| GPQA Diamond public · n=198 | 90.4 | $0.638 | 38.2 | LIVE_MEASURED |
| SWE-bench Verified public · n=500 | 74.2 | $0.695 | 36.5 | LIVE_MEASURED |
| Terminal Bench public · n=80 | 44.6 | $0.695 | 107.0 | LIVE_MEASURED |
| OSWorld public · n=369 | 38.2 | $0.695 | 104.9 | LIVE_MEASURED |
| FrontierMath public · n=300 | 48.8 | $0.638 | 81.3 | LIVE_MEASURED |
| LiveCodeBench public · n=612 | 82.1 | $0.695 | 37.3 | LIVE_MEASURED |
| MMMU public · n=1730 | 73.6 | $0.638 | 36.4 | LIVE_MEASURED |
| LongContext-128k public · n=120 | 82.8 | $0.638 | 37.4 | LIVE_MEASURED |
| GBU Private Rotating — Sep 2026 private · n=240 | 77.5 | $0.638 | 36.8 | LIVE_MEASURED |
| Agent Episodes public · n=96 | 66.2 | $0.695 | 114.1 | LIVE_MEASURED |
| Structured JSON public · n=400 | 95.1 | $0.638 | 38.6 | LIVE_MEASURED |
| CAD Science Lab public · n=64 | 81.4 | $0.638 | 37.2 | LIVE_MEASURED |