February 17, 2026

Cheaper Opus or open-source panic?

Claude Sonnet 4.6

Cheaper 'almost-Opus', mega memory, and comment chaos

TLDR: Anthropic’s Claude Sonnet 4.6 ships with a giant 1M-token memory and better “computer-using” skills at the same price. Commenters split between “cheaper near-Opus,” “not great for pure coding,” and “open-source-style rivals are catching up,” while many cheer the huge context—especially if it lands in Claude Code.

Claude Sonnet 4.6 just dropped, and the comment section lit up like a Friday night release. The pitch: bigger memory - a 1 million-token context that can hold whole codebases and contracts - plus sharper skills using a computer like a human, all at the same price. Anthropic claims near top-tier performance for many office tasks, and early devs say they prefer it to 4.5. One commenter summed up the mood with a wink: "Opus 4.6 but cheaper". Value play unlocked.

But the hot takes are spicy. Skeptics barked, "What happened to Sonnet 5?" while others went for the jugular: Anthropic is "running scared of open-weight models" and hyping safety before an IPO. The practical crowd countered with a shrug: might be worse than Opus for pure coding, but a win for people building assistants that click around apps for you. If that 1M context lands in Claude Code by default, devs say it's a game-changer.

Meanwhile, the "AI that uses a computer like you do" claim stirred debate. Anthropic cites OSWorld gains and better defenses against sneaky website instructions (see the docs). Fans joked it's a warm, funny spreadsheet intern; skeptics called it a therapy bot that tabs too much. Either way, the drama's premium - the price isn't.

Key Points

  • Anthropic released Claude Sonnet 4.6 with upgrades in coding, computer use, long-context reasoning, and agent planning.
  • Sonnet 4.6 introduces a beta 1M-token context window and becomes the default model for Free and Pro users in claude.ai and Claude Cowork.
  • Pricing is unchanged from Sonnet 4.5 at $3/$15 per million tokens.
  • OSWorld benchmarks show steady computer-use gains; early users report human-level capability on certain tasks, though the model still lags top human users.
  • Safety evaluations indicate Sonnet 4.6 is as safe as or safer than recent models, with improved resistance to prompt injection and user preference over Sonnet 4.5 (~70% in Claude Code).

Hottest takes

"Opus 4.6 but cheaper" — a_void_sky
"running scared of the open weight models" — rvz
"Probably not worth it for coding" — cube2222
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.