Anthropic appears to be quietly testing Opus 5.2 via Claude Code

On September 15, multiple users independently reported that Anthropic appears to be gray-scale testing a new model, Opus 5.2, through Claude Code: requests meant for Opus 5 were routed to a new model, and slugs explicitly referencing Opus 5.2 appeared in the code. No official announcement has been made, so the final naming and release timing remain unconfirmed.

Confirmed

• A slug explicitly referencing Opus 5.2 appeared in Claude Code, with Opus 5 requests routed to the new model, independently found by multiple users (@Angaisb, @koltregaskes, @CtrlAltDwayne)
• Side-by-side screenshots show behavioral differences for the same request (e.g., without search)
• @koltregaskes found the new model proactively iterates to fix issues across multiple rounds without prompting, resembling a Matt Shumer-style gauntlet loop
• @CtrlAltDwayne said the whole experience felt clearly different from Opus 5 — "even the way it talks" — near-Fable-style speed, improved design and visual reasoning, and extremely low usage consumption for the task
• 新智元 reported faster responses, crisper output, and autonomous "grueling loops" of sustained code iteration on long tasks

Unconfirmed

• Whether Anthropic skipped 5.1 to ship 5.2: some users speculate it skipped, while @CtrlAltDwayne referred to it as Opus 5.1/5.2
• Release timing: @CtrlAltDwayne guesses tomorrow or this week, purely personal speculation
• 新智元's claim that "Anthropic's internal Model2 now handles most code writing" is unverified
• Whether the perceived improvements are real or placebo: @CtrlAltDwayne admits the latter can't be ruled out

Why it matters

• If true, this is another silent rollout by Anthropic via Claude Code, letting users try a new model before any official announcement
• The traits observed — autonomous multi-round iteration, no slacking on long tasks, low token consumption — target core pain points of coding agents and could meaningfully affect developer workflows

2026-09-15 ~ 2026-09-15 · 6 related posts

• [source] Claude Opus 5.2 spotted in grayscale testing on Claude Code, seemingly skipping 5.1 — Angaisb_ · 2026-09-15
• [source] Claude Opus 5.2 spotted being tested inside Claude Code via model routing — koltregaskes · 2026-09-15
• Opus 5.2 spotted in Claude Code routing; leaked Model 2 reportedly writes most of Anthropic's code — 新智元 · 2026-09-15
• Rumor: Anthropic Soft-Launching Opus 5.1 Through Claude Code Before Full Rollout — CtrlAltDwayne · 2026-09-15
• Opus 5.1/5.2 spotted in Claude Code soft launch, release possibly this week — CtrlAltDwayne · 2026-09-15
• [source] Claude Code Users Speculate Hidden Opus 5.2 Upgrade Is Already Live — CtrlAltDwayne · 2026-09-15