Claude 4 Models

The Claude 4 generation began on May 22, 2025 with Claude Opus 4 and Claude Sonnet 4 (Introducing Claude 4, Anthropic, May 22, 2025) and ended a year later with Claude Opus 4.8 on May 28, 2026, before the Claude 5 family took over. In that year Anthropic cut Opus prices from $15 and $75 per million tokens to $5 and $25, grew the context window from 200K to 1M tokens, and introduced effort levels and adaptive thinking. As of Nov 6, 2026, six Claude 4 models remain on the Claude API as legacy models.

Quick facts

Sources: Pricing, Models overview and Model deprecations, Anthropic docs, read Nov 6, 2026.

Every Claude 4 model

Sources: Model deprecations and Context windows, Anthropic docs, read Nov 6, 2026. Haiku 4.5's retirement floor, Oct 15, 2026, has passed, but the docs still list it as active. Retired models may still run on Amazon Bedrock and Google Cloud, which set their own dates.

Opus 4 and Sonnet 4 (May 22, 2025)

Anthropic called both "hybrid models": near-instant answers, or extended thinking for harder problems, now able to use tools while thinking (beta). Anthropic reported 72.5% on SWE-bench Verified for Opus 4 and 72.7% for Sonnet 4, or 79.4% and 80.2% with parallel attempts. Opus 4 cost $15 and $75 per million tokens, Sonnet 4 $3 and $15, and Free users got Sonnet 4. Claude Code became generally available the same day, with VS Code and JetBrains extensions in beta. The post mentions protections for higher AI Safety Levels such as ASL-3 under Anthropic's Responsible Scaling Policy; the later Haiku 4.5 post confirms that Sonnet 4.5 and Opus 4.1 ran under ASL-3.

Opus 4.1 (Aug 5, 2025)

An upgrade on agentic tasks, coding and reasoning at the same price as Opus 4, with 74.5% on SWE-bench Verified without extended thinking.

Sonnet 4.5 (Sep 29, 2025)

Anthropic called it "the best coding model in the world" (its claim at the time) and said it could stay on complex tasks for more than 30 hours. It reported 77.2% on SWE-bench Verified and 61.4% on OSWorld, against Sonnet 4's 42.2% four months earlier. It launched with the Claude Agent SDK, Claude Code Checkpoints and a VS Code extension. It is deprecated and its replacement is claude-sonnet-5-5.

4:51 Anthropic's own engineers later described a quirk of Sonnet 4.5 that they called "context anxiety": it started wrapping up tasks early while it still had room left in its context window, so they added mitigations to their agent harness, and the behavior went away with Opus 4.5 (Ship your first Managed Agent, Claude, May 26, 2026).

Haiku 4.5 (Oct 15, 2025)

Coding close to Sonnet 4 "at one-third the cost and more than twice the speed", in Anthropic's words, at $1 and $5 per million tokens, with 73.3% on SWE-bench Verified. It was released under ASL-2, a lower safety level than its larger siblings. Among models still on the API, only Haiku 4.5 and Sonnet 4.5 lack effort control, and Haiku 4.5 has no adaptive thinking (Effort, Anthropic docs, read Nov 6, 2026). Its successor is Claude Haiku 5.5, at a tenth of the price for most prompts.

Opus 4.5 (Nov 24, 2025)

Opus 4.5 cut Opus pricing by two thirds to $5 and $25 and introduced the effort parameter: at medium effort it matched Sonnet 4.5's best SWE-bench Verified score with 76% fewer output tokens. Anthropic called it its "most robustly aligned model" to that date, and removed separate Opus caps in the apps. On the public SWE-bench site, Opus 4.5 at high effort tops the same-harness comparison at 76.8% on Verified (data read Nov 6, 2026; the site has no later Claude entries).

Opus 4.6 and Sonnet 4.6 (Feb 2026)

Opus 4.6 (Feb 5, 2026) was the first Opus with a 1M-token context window, 128K output, adaptive thinking and the max effort level, and the first model with a dateless pinned id. Sonnet 4.6 (Feb 17, 2026) became the default for Free and Pro in the apps, scored 80.2% on SWE-bench Verified with a prompt change, and stayed at $3 and $15. Both launched with 1M context in beta; it left beta at standard pricing on Mar 13, 2026. Old-style extended thinking with budget_tokens still works on both but is deprecated (Extended and Adaptive Thinking).

Opus 4.7 and Opus 4.8 (spring 2026)

Opus 4.7 (Apr 16, 2026) added a newer tokenizer that turns the same text into about 1.0 to 1.35 times as many tokens, image input up to 2,576 pixels on the long edge, and the xhigh effort level, which became Claude Code's default. Anthropic kept its cyber abilities "deliberately" below Mythos Preview.

0:52 Opus 4.8 (May 28, 2026) was, in Anthropic's own words, "a modest but tangible improvement", with fast mode at $10 and $50 and effort control brought to claude.ai and Cowork on every plan. It now serves as the fallback model when safeguards flag a request to Fable 5, Fable 5.1 or Opus 5, and most cybersecurity tasks sent to Claude Opus 5.5 are rerouted to it. Anthropic's docs also name it the replacement for retired Opus 4 and 4.1.

What reviewers found

2:52 In an Aug 12, 2026 comparison of the $20 plans, Parker Prompts put Claude Opus 4.8 at the top of the main independent intelligence ranking, just ahead of GPT-5.5, and said it led on coding and invented less on long, complicated answers; the channel gave Claude the raw-intelligence round "but only just".

7:41 In an Aug 26, 2026 video comparing four assistants, Peter Yang said Opus 4.6 had been the best Claude model for personality, and that Opus 5, which followed the 4 generation, felt "a little bit too judgmental" to him, so he had moved his everyday chats to ChatGPT. That is one person's opinion of tone, not a measurement.

Frequently asked questions

Is Claude Opus 4 still available? No. Claude Opus 4 and Sonnet 4 were retired on the Claude API on Jun 15, 2026, and Opus 4.1 on Aug 5, 2026. Some remain on Google Cloud or Amazon Bedrock, and Anthropic's docs recommend Claude Opus 4.8 as the replacement.

When does Claude Sonnet 4.5 retire? Claude Sonnet 4.5 retires on the Claude API on Nov 30, 2026, and requests to it then fail. Anthropic's recommended replacement is Claude Sonnet 5.5.

How much does Claude Opus 4.8 cost? Claude Opus 4.8 costs $5 per million input tokens and $25 per million output tokens on the Claude API, the same as Opus 4.5, 4.6, 4.7 and Opus 5, and $10 and $50 in fast mode.

Which Claude 4 model has a 1M context window? Claude Opus 4.6, Sonnet 4.6, Opus 4.7 and Opus 4.8 accept 1M tokens on the API. Opus 4.5, Sonnet 4.5 and Haiku 4.5 are limited to 200K.

Should I still use Claude 4 models? Anthropic recommends moving to the Claude 5.5 models: Claude Opus 5.5 is cheaper than any Claude 4 Opus at $4 and $20, and Claude Haiku 5.5 is cheaper than Haiku 4.5 for prompts up to 100,000 tokens.

Sources

• Ship your first Managed Agent, Claude, May 26, 2026 (video)
• Introducing Claude Fable 5, Anthropic, Jun 9, 2026 (video)
• ChatGPT vs Claude vs Gemini: Which $20 AI Plan Is Worth It?, Parker Prompts, Aug 12, 2026 (video)
• ChatGPT vs Claude vs Grok vs Gemini: The Best AI for 10 Use Cases (August 2026), Peter Yang, Aug 26, 2026 (video)
• Introducing Claude 4, Anthropic, May 22, 2025
• Claude Opus 4.1, Anthropic, Aug 5, 2025
• Introducing Claude Sonnet 4.5, Anthropic, Sep 29, 2025
• Introducing Claude Haiku 4.5, Anthropic, Oct 15, 2025
• Introducing Claude Opus 4.5, Anthropic, Nov 24, 2025
• Introducing Claude Opus 4.6, Anthropic, Feb 5, 2026
• Introducing Claude Sonnet 4.6, Anthropic, Feb 17, 2026
• Introducing Claude Opus 4.7, Anthropic, Apr 16, 2026
• Introducing Claude Opus 4.8, Anthropic, May 28, 2026
• Models overview, Anthropic docs, read Nov 6, 2026
• Pricing, Anthropic docs, read Nov 6, 2026
• Model deprecations, Anthropic docs, read Nov 6, 2026
• Context windows, Anthropic docs, read Nov 6, 2026
• Effort, Anthropic docs, read Nov 6, 2026
• Release notes, Anthropic docs, read Nov 6, 2026
• SWE-bench leaderboards data, SWE-bench, read Nov 6, 2026