Reasoning Models
Reasoning models are trained or prompted to spend extra inference-time ‘thinking’ before answering — the o1/o3-style and R1-style models that trade latency and cost for materially better performance on math, coding, and multi-step logic problems that trip up a standard chat model. This category emerged as a distinct model class rather than just a prompting technique, with dedicated training recipes and pricing tiers. GROUNDING tracks new reasoning-model releases and the benchmarks (math, competitive coding, agentic tool-use) where this category’s advantage over general chat models is largest and most builder-relevant.
At a glance
- 13 tracked models (1 hand-curated)
- Most recently updated: OpenAI-o1 (2026-07-06)
Most actively covered
- DeepSeek R1 ★
- o1
- R1
- o3
- Gemini 3.1 Pro Thinking
- OpenAI-o1
- DeepSeek R1 8B
- DeepSeek-R1-Distill-Qwen-1.5B
- gpt-5.5-thinking
- o3-mini
FAQ
What is the Reasoning Models category?
Models optimized for extended step-by-step reasoning (o1/o3/R1-style).
How many models are in this category?
GROUNDING currently tracks 13 models in Reasoning Models, 1 of which have a hand-curated profile.
How current is this hub?
The most recently updated entry is OpenAI-o1, last mentioned 2026-07-06; the radar refreshes hourly.