Skip to content

Claude Sonnet 4.7: What's Actually Out and What's Still a Rumor

Shivam Malani
Claude Sonnet 4.7: What's Actually Out and What's Still a Rumor

If you came looking for Claude Sonnet 4.7, here's the blunt truth: Anthropic hasn't released it. The model that actually shipped on April 16, 2026 is Claude Opus 4.7, the new flagship in the Claude 4 family. Sonnet 4.7 has floated around in leaks and forum chatter for months, but the current Sonnet on the API is still Sonnet 4.6.

Quick answer: Claude Sonnet 4.7 is not a released product. The latest Sonnet available through Anthropic is Claude Sonnet 4.6. The newest model Anthropic has launched is Claude Opus 4.7.

What Anthropic actually shipped

Claude Opus 4.7 rolled out on April 16, 2026 across claude.ai, the Claude Platform, Amazon Bedrock, Google Cloud's Vertex AI, and Microsoft Foundry. It's positioned as the most capable generally available Claude model, with a step-change improvement in agentic coding over Opus 4.6, a 1M-token context window, and a new adaptive thinking mode that decides how long to reason based on task complexity.

Pricing lands at $5 per million input tokens and $25 per million output tokens, matching Opus 4.6. Prompt caching cuts costs up to 90%, batch processing saves 50%, and US-only inference is available at a 1.1x premium. The API model ID is claude-opus-4-7.


The current Claude lineup

ModelAPI IDContextPricing (in / out per MTok)
Claude Opus 4.7claude-opus-4-71M tokens$5 / $25
Claude Sonnet 4.6claude-sonnet-4-61M tokens$3 / $15
Claude Haiku 4.5claude-haiku-4-5200k tokens$1 / $5

Sonnet 4.6 supports both extended thinking and adaptive thinking. Opus 4.7 uses adaptive thinking only. Haiku 4.5 supports extended thinking but not adaptive thinking. All three handle text and image input, with multilingual and vision capabilities.


Where the "Sonnet 4.7" idea came from

The name has been circulating since late 2025, driven by leaks and speculation inside the Claude community. A leaked Claude Code source map in late March 2026 included an internal "Undercover Mode" feature with a blocklist that reportedly referenced unreleased version strings, including Sonnet 4.8 and Opus 4.7. Notably, that blocklist pointed to Sonnet 4.8, not 4.7, which makes the "Sonnet 4.7" label itself look more like community shorthand than a confirmed codename.

Anthropic has used a 0.5-step cadence for Sonnet so far (Sonnet 4, Sonnet 4.5, Sonnet 4.6), with Sonnet 3.7 being a one-time naming quirk during the Claude 3 era. There's no public confirmation that the next Sonnet will be numbered 4.7 rather than 4.7's neighbors.


What Opus 4.7 changes if you were waiting on a Sonnet upgrade

Opus 4.7's release note highlights three shifts that matter even if you stay on Sonnet 4.6:

  • Updated tokenizer: The same input maps to roughly 1.0 to 1.35x more tokens depending on content type. This applies to Opus 4.7 specifically and affects cost calculations.
  • Adaptive thinking by default: On claude.ai, Opus 4.7 removes the manual extended thinking toggle and decides effort automatically. Claude Code still exposes effort levels (low, medium, high, xhigh, max).
  • Stricter instruction following: Prompts tuned loosely for Opus 4.6 may now be interpreted more literally. Re-test any skill files, system prompts, or agent scaffolds.

Sonnet 4.6 remains the balanced option. It keeps the 1M context window, costs 40% of Opus 4.7 on input and 60% on output, and retains both extended and adaptive thinking.


Known regressions in Opus 4.7

The Opus 4.7 system card documents a meaningful drop on MRCR v2 at 1M tokens, falling from 78.3% on Opus 4.6 to 32.2% on Opus 4.7. Anthropic's engineering lead on Claude Code, Boris Cherny, explained that MRCR is being phased out because it stacks distractors in ways that don't reflect real usage, and that GraphWalks is now the preferred long-context signal.

The model card also flags specific failure modes: occasional hallucinated document quotes, overconfident initial diagnoses on technical problems, and occasional misreporting of test failures as pre-existing. If precise ordinal retrieval from long documents is central to your workflow, Sonnet 4.6 or Opus 4.6 may still be the better fit.


How to pick the right model right now

Use caseRecommended model
Frontier coding, long agent runs, complex enterprise workflowsClaude Opus 4.7
Balanced coding and reasoning at lower cost, long documents with precise retrievalClaude Sonnet 4.6
High-volume, latency-sensitive, cheap tool calls and formattingClaude Haiku 4.5
Anything requiring a model literally called "Sonnet 4.7"Not available

If you're writing code against the API

Use the current, stable model IDs. There is no claude-sonnet-4-7 endpoint.


# Opus 4.7 (latest flagship)
claude-opus-4-7

# Sonnet 4.6 (latest Sonnet)
claude-sonnet-4-6

# Haiku 4.5
claude-haiku-4-5

Valid Claude model IDs as of April 2026

Pointing an application at a nonexistent claude-sonnet-4-7 ID will return an error. If you're migrating from Opus 4.6 or earlier, the Claude API docs cover the current model aliases and capabilities.


What about Mythos?

Anthropic's Opus 4.7 benchmark charts compare against Claude Mythos Preview, a separate research-preview model tied to Project Glasswing for defensive cybersecurity workflows. Mythos is invitation-only, not self-serve, and not a successor to Sonnet. Treat it as a parallel research track, not a consumer or general developer model.


The short version: if your roadmap depends on "Claude Sonnet 4.7," it depends on a model that doesn't exist yet. Anthropic hasn't confirmed a next-generation Sonnet release window. For now, Opus 4.7 is the new top of the lineup, Sonnet 4.6 is the current mid-tier workhorse, and anything beyond that is speculation.