Claude models, compared.

The history of Claude.

Every Claude release since 2023, newest first, each linked to Anthropic's own announcement.

1. Jun 30, 2026
   Claude Sonnet 5

   • The most agentic Sonnet yet, close to Opus 4.8 at lower prices.
   • Became the default model on Free and Pro.
   • Launched at $2 input / $10 output per MTok introductory pricing.

   announcement
2. Jun 9, 2026
   Claude Fable 5 + Claude Mythos 5

   • The Mythos-class debut, a tier above Opus.
   • Fable 5 is the Mythos-class model made safe for general availability.
   • Briefly paused under a US government order, redeployed from Jul 1, 2026.
   • Mythos 5 runs with the US government under Project Glasswing.

   announcement
3. May 28, 2026
   Claude Opus 4.8

   • Fast mode runs at 2.5x speed, three times cheaper than previous fast modes.
   • Claude Code gained dynamic workflows running hundreds of parallel subagents.

   announcement
4. Apr 16, 2026
   Claude Opus 4.7

   • The tokenizer changed: the same input can map to roughly 1.0 to 1.35x more tokens depending on content.
   • Shipped higher-resolution vision.

   announcement
5. Feb 17, 2026
   Claude Sonnet 4.6

   • Brought 1M context to the Sonnet tier.
   • Testers preferred it to Sonnet 4.5 about 70% of the time, and even to Opus 4.5 about 59% of the time.

   announcement
6. Feb 5, 2026
   Claude Opus 4.6

   • Brought 1M context in beta.
   • Agent teams landed in Claude Code.
   • Topped Terminal-Bench 2.0.

   announcement
7. Nov 24, 2025
   Claude Opus 4.5

   • The big Opus price drop, to $5 input / $25 output per MTok.
   • Scored higher than any human candidate ever on Anthropic's internal 2-hour engineering exam.

   announcement
8. Oct 15, 2025
   Claude Haiku 4.5

   • Near-Sonnet-4 coding at a third of the cost and more than twice the speed.
   • Priced at $1 input / $5 output per MTok.

   announcement
9. Sep 29, 2025
   Claude Sonnet 4.5

   • Hit 77.2% on SWE-bench Verified.
   • Observed holding focus for more than 30 hours on multi-step tasks.
   • The Claude Agent SDK shipped alongside it.

   announcement
10. Aug 5, 2025
    Claude Opus 4.1

    • Pushed SWE-bench Verified to 74.5%.

    announcement
11. May 22, 2025
    Claude 4: Opus 4 + Sonnet 4

    • Anthropic's flat claim: the world's best coding model.
    • A customer ran a 7-hour autonomous refactor.
    • Claude Code went generally available.

    announcement
12. Feb 24, 2025
    Claude 3.7 Sonnet

    • The first hybrid reasoning model on the market.
    • Launched alongside Claude Code.
    • Cut unnecessary refusals by 45%.

    announcement
13. Oct 22, 2024
    Upgraded 3.5 Sonnet + Claude 3.5 Haiku

    • The first frontier model with computer use in public beta.
    • Lifted SWE-bench Verified from 33.4% to 49.0%.
    • The new Haiku matched Claude 3 Opus.

    announcement
14. Jun 21, 2024
    Claude 3.5 Sonnet

    • The coding inflection: a mid-tier model that beat Claude 3 Opus at twice the speed.
    • Solved 64% of internal agentic coding problems versus Opus's 38%.
    • Shipped Artifacts.

    announcement
15. Mar 4, 2024
    Claude 3: Opus, Sonnet, Haiku

    • The release that invented the Opus, Sonnet, Haiku naming and the first Opus tier.
    • The first Claude to outperform its peers on most common benchmarks.
    • Near-perfect, 99%+, needle-in-a-haystack recall.

    announcement
16. Nov 21, 2023
    Claude 2.1

    • Brought a 200K context window, over 500 pages.
    • Halved false statements versus Claude 2.0.
    • Tool use arrived in beta.

    announcement
17. Jul 11, 2023
    Claude 2

    • The first public claude.ai chat site, in US and UK beta.
    • 100K context, hundreds of pages in one prompt.
    • Coding jumped to 71.2% on Codex HumanEval, up from 56.0%.

    announcement
18. Mar 14, 2023
    Claude and Claude Instant

    • The first public Claude, in two tiers.
    • Came out of closed alpha with partners like Notion, Quora, and DuckDuckGo.

    announcement