Claude 1

// Overview

Claude 1 was Anthropic's first publicly released large language model, launched on March 14, 2023 alongside its faster, cheaper sibling Claude Instant when Anthropic opened the Anthropic API for early access. It shipped in two tiers: the high-performance Claude (model IDs claude-1.0 through claude-1.3, sometimes aliased claude-v1) for tasks needing complex reasoning, and the low-latency Claude Instant. Anthropic described Claude as an assistant for summarization, search, creative and collaborative writing, Q&A, coding, and other text tasks, and emphasized that early customers found it less likely to produce harmful outputs, easier to converse with, and more steerable.

What set Claude 1 apart was its training method. Where most assistants of the era relied purely on human feedback, Claude was trained with Constitutional AI — a technique Anthropic published in December 2022 that uses a written set of principles (a 'constitution') and AI-generated feedback to steer the model toward being helpful, honest, and harmless. That approach, built to reduce reliance on human labelers for harmful-content review, became the foundation for every later Claude model. The 'Claude' name is a nod to the field's heritage, evoking information-theory pioneer Claude Shannon.

Claude 1's most consequential feature arrived after launch. The models started with a 9,000-token context window, but on May 11, 2023 Anthropic expanded Claude and Claude Instant to 100,000 tokens — roughly 75,000 words — one of the first frontier-lab models able to read book-length documents in a single prompt. The most capable point release, claude-1.3, set the baseline that Claude 2 (July 2023) measured itself against. The entire Claude 1 and Claude Instant 1.x line was deprecated on September 4, 2024 and retired on November 6, 2024; it is no longer callable.

// Benchmarks

1. Bar exam (multiple-choice section), claude-1.373%
2. GSM8K (grade-school math), claude-1.385.2%
3. Codex HumanEval (Python coding), claude-1.356%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

Original 2023 pay-as-you-go API pricing for the flagship Claude (claude-v1). The cheaper Claude Instant tier was $1.63/M prompt and $5.51/M completion — about one-sixth the cost. These models are retired and no longer sold.

Pricing source ↗

// Strengths

• Constitutional AI training (RLHF + RLAIF) made it notably steerable and less prone to harmful or evasive outputs than peer assistants of early 2023
• Long-context pioneer: among the first frontier models to offer a 100K-token window (May 2023), enabling whole-document and book-length analysis
• Strong at summarization, Q&A, and long-form / collaborative writing — Anthropic's headline launch use cases
• Two-tier lineup let developers trade off cost and latency: flagship Claude for hard reasoning, Claude Instant at roughly one-sixth the price for high-throughput work

// Best for

• Summarizing long documents, transcripts, and reports
• Question answering and search over provided text
• Creative and collaborative long-form writing and editing
• Drafting and explaining code
• High-volume, latency-sensitive text processing via the cheaper Claude Instant tier

// How to access

// Claude (legacy 1–2.x) — every version

The full lineage of the Claude (legacy 1–2.x) line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.