Every Claude Model Ever Made: The Complete History from Claude 1 to Opus 4.6

Anthropic launched Claude 1 in March 2023 as a direct alternative to GPT-3.5, designed from the start with safety as a first-class concern. The initial release offered a 9,000-token context window, enough for substantial documents but limited compared to what would follow. In May 2023, Anthropic shipped Claude 1.3, expanding context to 100,000 tokens. That milestone let users load entire codebases, legal contracts, and book-length documents into a single conversation, opening Claude to a new class of enterprise use cases almost immediately.

Claude 2 launched on July 11, 2023. It came with sharply improved coding and reasoning, real benchmark scores to back the claim, and public access through a new web interface at claude.ai, initially available for US and UK users. On the Bar Exam, Claude 2 scored 76.5%, up from Claude 1.3's 73.0%. On the Codex HumanEval Python coding benchmark, it reached 71.2%, up from 56.0%. On the GSM8k math benchmark, it scored 88.0%, up from 85.2%. GRE Reading and Writing performance placed Claude 2 above the 90th percentile of human test takers. Internal red-teaming showed Claude 2 produced twice as many harmless responses as Claude 1.3 when pushed on sensitive topics. API pricing matched Claude 1.3 rates, so customers got significantly more capability at no additional cost. In November 2023, Claude 2.1 expanded the context window to 200,000 tokens and cut hallucination rates substantially.

The Claude 3 family launched on March 4, 2024, with three models positioned across different price and performance points: Haiku, Sonnet, and Opus. All three launched with 200,000-token context windows, and select customers could access contexts exceeding 1 million tokens. Every Claude 3 model added vision, enabling the processing of photos, charts, diagrams, and other visual content alongside text.

In June 2024, Anthropic released Claude 3.5 Sonnet. It outperformed Claude 3 Opus on GPQA, MMLU, and HumanEval benchmarks while running at twice Opus's speed, at Sonnet-tier pricing of $3 per million input tokens and $15 per million output tokens. On an internal agentic coding evaluation, Claude 3.5 Sonnet solved 64% of problems, compared to Claude 3 Opus's 38%. The release also introduced Artifacts, a dedicated workspace that lets users view, edit, and build on AI-generated code, websites, and documents alongside the conversation.

The Claude 4 generation shifted focus toward long-horizon agentic work: complex, multi-step tasks where the model takes autonomous actions without constant human intervention. Models in this series introduced adaptive thinking, which applies reasoning effort proportionally to task complexity rather than always engaging full extended thinking.

Pricing dropped significantly across the Claude 4 generation. Claude Opus 4.1 started at $15 per million input tokens and $75 per million output tokens, a structure carried over from Claude 3 Opus. By the time Claude Opus 4.6, 4.7, and 4.8 arrived, pricing settled at $5 per million input tokens and $25 per million output tokens, while capability increased substantially. Context windows expanded to 1 million tokens across the Opus tier, with maximum output of 128,000 tokens. Claude Sonnet 4.6 (at $3 input and $15 output per million tokens) and Claude Haiku 4.5 (at $1 input and $5 output per million tokens) extended the tiered structure downward for speed-sensitive and cost-sensitive workloads.

Starting with Claude Opus 4.7, Anthropic introduced a new tokenizer. The same input text produces roughly 30% more tokens compared to models before Claude Opus 4.7, which affects cost calculations when migrating between model generations.

Claude Opus 4.1 is deprecated and scheduled for retirement on August 5, 2026. Teams still running on Claude Opus 4.1 should migrate to Claude Opus 4.8 before that date.

On June 9, 2026, Anthropic launched Claude Fable 5 and Claude Mythos 5, the fifth-generation frontier models. Both offer 1 million token context windows, 128,000 token maximum output, adaptive thinking enabled by default, and pricing at $10 per million input tokens and $50 per million output tokens.

Fable 5 is Anthropic's most capable widely released model and is generally available through the Claude API, Amazon Bedrock, Vertex AI, and Microsoft Foundry. Mythos 5 is not generally available: it is offered through Project Glasswing in limited access to approved customers focused on defensive cybersecurity workflows. To request access, contact your Anthropic, AWS, or Google Cloud account team.