Models
GPT-5
| Provider | OpenAI |
|---|---|
| Context Window | 128K tokens |
| Parameters | Unknown |
| Input Pricing | $10/1M |
| Output Pricing | $30/1M |
| API Available | ✓ Yes |
| Open Source | ✗ No |
GPT-5 is OpenAI’s latest flagship model, released in April 2026. It represents a significant leap from GPT-4o, particularly in abstract reasoning, math, and instruction adherence.
Capabilities
- Reasoning: Leads the ARC-AGI benchmark at 97.2% — the highest ever recorded
- Code: 98.1% on HumanEval, best-in-class for competitive programming
- Math: 96.3% on MATH benchmark
- Multimodal: Vision, audio transcription, image analysis
- Tool use: Reliable function calling with parallel tool execution
Performance Summary
On MMLU: 94.8% | On HumanEval: 98.1% | On MATH: 96.3% | On ARC-AGI: 97.2%
Best For
Tasks requiring the absolute highest accuracy — complex legal/financial analysis, advanced mathematics, competitive programming, novel scientific reasoning.
Context Window Notes
GPT-5 uses a 128K context window, smaller than Claude’s 200K but still sufficient for most enterprise workloads. Token recall quality is excellent throughout the full window.