Agentic Reasoning (o1, o3, Reasoning Models)
Models that 'think' before answering — internal chains-of-thought, test-time compute, and self-correction.
Agentic Reasoning (o1, o3, Reasoning Models) is a free AI Agents lesson on CoddyKit — lesson 1 of 4. You can read the complete lesson below for free — then practise it hands-on in the browser with a built-in code editor and a 24/7 AI tutor. It is part of the AI Agents learning path, one of 4 lessons in the course, and your progress syncs across the web and the CoddyKit app.
A New Class of Models
OpenAI o1 (Sep 2024) introduced "reasoning models" — LLMs that explicitly reason step-by-step before producing the visible answer. Successors include o3, DeepSeek R1, Claude with extended thinking, Gemini 2.5 thinking, and many others.
How They Work
Instead of generating the answer immediately, the model:
- Produces a long internal "thinking" sequence (CoT-style)
- Explores multiple approaches
- Self-corrects
- Then emits the final visible answer
Test-Time Compute
You can choose how much "thinking" the model does. More tokens spent thinking = better answers on hard problems. The trade-off is latency and cost.
OpenAI o-Series API
response = client.chat.completions.create(
model='o3-mini',
messages=[{'role': 'user', 'content': 'Solve this puzzle...'}],
reasoning_effort='high' # low / medium / high
)
print(response.choices[0].message.content)
print(response.usage.reasoning_tokens) # how many 'thinking' tokens were usedAnthropic Extended Thinking
response = client.messages.create(
model='claude-sonnet-4-5',
max_tokens=8192,
thinking={'type': 'enabled', 'budget_tokens': 4096},
messages=[{'role': 'user', 'content': 'Hard reasoning problem'}]
)
# response.content includes 'thinking' blocks + 'text' final answerWhen Reasoning Models Shine
- Multi-step math
- Code generation with complex logic
- Legal / scientific reasoning
- Planning that needs lookahead
Bench: o3 scores >90% on AIME and high on GPQA — leagues above non-reasoning models.
When Standard Models Are Better
- Simple chat / Q&A — reasoning is wasted
- Latency-critical paths — reasoning adds seconds
- Cost-sensitive bulk work — reasoning multiplies cost
- Style / formatting tasks — no benefit
Cost Profile
Reasoning tokens count in pricing. o3-mini at high effort can use 10k-100k thinking tokens per question. That can be $0.10 - $1 per question on the bigger reasoning models.
Inside Agent Loops
Using a reasoning model as the planner inside a larger agent:
- Reasoning model: plan / decide / evaluate
- Standard model: execute steps
- Tool calls: run normally
Best of both: deep reasoning where it matters, cheap execution elsewhere.
Don't Force CoT on Top
You no longer need "let's think step by step" prompts with reasoning models — they already do CoT internally. Adding it can actually hurt.
Open-Source Reasoning
DeepSeek R1, R1-Distill, and others bring reasoning to open weights. Available on HuggingFace, Together AI, etc.
Research Frontier
- Reflection-tuning — train models to critique themselves
- Search-based inference — Tree-of-Thoughts, MCTS at inference time
- Test-time training — adapt weights per query
Caveat: Black-Box Thinking
Reasoning model "thoughts" are usually hidden from you (OpenAI) or summarised (Anthropic). Limited transparency makes debugging harder; rely on input/output evals instead.
When to Use Reasoning Models
For which task is a reasoning model most worth the extra cost?
Recap
Reasoning models (o1, o3, R1, Claude extended thinking) trade time/cost for quality on hard tasks. Use as planner/judge inside agents. Skip for simple chat.
Frequently asked questions
Is the “Agentic Reasoning (o1, o3, Reasoning Models)” lesson free?
Yes — the full text of “Agentic Reasoning (o1, o3, Reasoning Models)” is free to read here on the web, and the AI Agents course includes 4 lessons in total. To practise it interactively (a built-in code editor and a 24/7 AI tutor) and unlock the rest of the AI Agents course, upgrade to CoddyKit PRO.
What will I learn in “Agentic Reasoning (o1, o3, Reasoning Models)”?
Models that 'think' before answering — internal chains-of-thought, test-time compute, and self-correction. You practise AI Agents with hands-on code you run directly in the browser, and a 24/7 AI tutor answers your questions as you work through the lesson.
Do I need any experience to start AI Agents?
No prior experience is required. AI Agents on CoddyKit is structured for beginners through advanced learners; this is — lesson 1 of 4, so you can start here or from the beginning and move at your own pace.
How long does the “Agentic Reasoning (o1, o3, Reasoning Models)” lesson take?
Most CoddyKit lessons take about 5–10 minutes. Each one is bite-sized and interactive, so you make steady progress and pick up exactly where you left off across the web and the app.
Can I write and run code in this AI Agents lesson?
Yes. Every AI Agents lesson includes a built-in code editor, so you write and run real code right in your browser and get instant AI feedback — no local setup required.
All lessons in this course
- Agentic Reasoning (o1, o3, Reasoning Models)
- Hybrid Symbolic + Neural Agents
- Multimodal Agents (Vision + Voice + Action)
- Open Problems: Robustness, Alignment, Long-Horizon Memory