Claude is a family of large language models developed by Anthropic, founded in 2021 by former OpenAI researchers including Dario Amodei and Ilya Sutskever's colleagues. The Claude 3 family (Haiku, Sonnet, Opus) launched in March 2024 and became the first models to seriously challenge GPT-4 on reasoning and coding benchmarks. Claude 3.5 Sonnet, released June 2024, set a new state-of-the-art on SWE-Bench coding, outperforming every model including GPT-4o.
Constitutional AI: how Anthropic builds safety in
Claude is trained using Constitutional AI (CAI), a technique Anthropic invented that replaces human preference labelers with a written 'constitution' — a set of principles the model uses to evaluate and revise its own outputs. This allows Anthropic to scale alignment without requiring humans to label millions of harmful examples.
- Step 1 — SL-CAI: Model generates responses to red-team prompts, then critiques and revises its own outputs using the constitution
- Step 2 — RL-CAI: A preference model (PM) is trained on the AI's own critiques, then used as a reward signal for RLHF
- The constitution includes principles from the UN Declaration of Human Rights, Apple ToS, Anthropic's own guidelines
- This makes Claude's refusals more principled and consistent — it can explain *why* it won't do something, not just refuse
CAI vs standard RLHF
Standard RLHF relies entirely on human raters to signal what is harmful. Constitutional AI augments this with self-critique — the model learns to evaluate its own outputs, enabling more consistent alignment at scale.
Claude model family and capabilities
| Model | Released | Best for | Context | Speed |
|---|---|---|---|---|
| Claude 3 Haiku | Mar 2024 | Fast, cheap tasks | 200K tokens | Fastest |
| Claude 3 Sonnet | Mar 2024 | Balanced | 200K tokens | Fast |
| Claude 3 Opus | Mar 2024 | Complex reasoning | 200K tokens | Slower |
| Claude 3.5 Sonnet | Jun 2024 | Coding, analysis | 200K tokens | Fast |
| Claude 3.5 Haiku | Nov 2024 | Production speed | 200K tokens | Fastest |
| Claude 3.7 Sonnet | Feb 2025 | Extended thinking | 200K tokens | Fast |
Claude's 200K context window (roughly 150,000 words — the length of an entire novel) was a major breakthrough when released. It allows Claude to analyze entire codebases, legal documents, or research papers in a single prompt. GPT-4 had a 128K context window; Gemini 1.5 Pro later extended this to 1 million tokens.
Claude's 'extended thinking' mode
Claude 3.7 Sonnet introduced 'extended thinking' — a mode where the model explicitly reasons through a problem step-by-step before producing a final answer, similar to OpenAI's o1 reasoning model. This is implemented as a separate thinking token budget: you can allocate up to 128K tokens for internal reasoning, after which the model produces its final visible response.
Using Claude with extended thinking
import anthropic
client = anthropic.Anthropic()
response = client.messages.create(
model="claude-3-7-sonnet-20250219",
max_tokens=16000,
thinking={
"type": "enabled",
"budget_tokens": 10000 # tokens for internal reasoning
},
messages=[{
"role": "user",
"content": "Prove that the square root of 2 is irrational."
}]
)
# Response has two blocks: thinking + answer
for block in response.content:
if block.type == "thinking":
print("REASONING:", block.thinking[:200])
elif block.type == "text":
print("ANSWER:", block.text)On LumiChats
LumiChats gives you access to the full Claude 3.5 and 3.7 model family — including extended thinking mode — alongside GPT-4o and Gemini in one interface.
Try it free