GPT‐5 - chunhualiao/public-docs GitHub Wiki
| Feature | GPT-5 (Standard) | GPT-5 Thinking | GPT-5 Pro |
|---|---|---|---|
| Purpose | General-purpose AI with automatic mode switching | Deep reasoning mode for complex tasks | Maximum performance for advanced reasoning |
| Availability | All users (Free, Plus, Pro, Team) | Auto-activated or manual (Plus, Pro, Team) | Pro/Team subscribers only |
| Speed | Very fast (optimized for quick responses) | Slower (prioritizes depth over speed) | Slowest (uses extended compute for rigor) |
| Reasoning Depth | Low to medium (auto-routes to Thinking as needed) | High (multi-step, chain-of-thought reasoning) | Maximal (parallel test time compute) |
| Accuracy | ~45% fewer factual errors vs. GPT-4o | ~80% fewer factual errors vs. o3 | >90% fewer major errors vs. o3 |
| Coding Performance | 52.8% (SWE-bench, without Thinking) | 74.9% (SWE-bench, with Thinking) | Higher accuracy, fewer logic errors |
| Math Performance | 93.3% (HMMT, no tools) | 96.7% (HMMT, with Python) | 100% (HMMT, with Python) |
| Science Performance | 77.8% (GPQA Diamond, no tools) | 85.7% (GPQA Diamond, with Thinking) | 89.4% (GPQA Diamond, with Python) |
| Use Cases | Everyday Q&A, writing, summaries | Complex coding, technical analysis, strategies | Research-grade tasks, large datasets, critical work |
| Context Window | Up to 400K tokens | Up to 400K tokens | Up to 400K tokens |
| Output Window | Up to 128K tokens | Up to 128K tokens | Up to 128K tokens |
| API Pricing | $1.25/$10 per 1M tokens (input/output) | Same as GPT-5 | Same as GPT-5 (Pro plan required) |
| Hallucination Rate | Moderate reduction vs. GPT-4o | Significant reduction (~4.8% error rate) | Lowest (~1.6% on HealthBench) |
| Tool Integration | Basic (auto-routed as needed) | Strong (sequential/parallel tool calls) | Advanced (Deep Research, custom connectors) |
Notes:
- GPT-5 (Standard): Default model in ChatGPT, uses a real-time router to switch between fast responses and deeper reasoning based on query complexity.
- GPT-5 Thinking: Automatically or manually engaged for tasks requiring multi-step reasoning, such as coding or strategic planning.
- GPT-5 Pro: Offers the highest accuracy and extended reasoning, ideal for mission-critical tasks like scientific research or large-scale codebases.
- Data sourced from various benchmarks and OpenAI's system card.