Claude Report — 2026-04-17
- Claude Opus 4.7 goes GA: 64.3% SWE-Bench Pro, 3.75MP high-res vision, new tokenizer (up to +35% token usage).
- Three breaking API changes in Opus 4.7: `temperature`/`top_p`/`top_k` now return 400; `budget_tokens` in thinking removed; thinking content hidden by default.
- Claude Code v2.1.111: `/ultrareview` (parallel multi-agent code review), `/less-permission-prompts` skill, Windows PowerShell tool rollout.
- Sonnet 4 and Opus 4 deprecated; retire June 15, 2026 — migrate to Sonnet 4.6 / Opus 4.7.
- HN thread (1,759 pts, 1,268 comments): developers flag adaptive thinking regressions and hidden reasoning output as first-day concerns.
🚀 Anthropic Official
Claude Opus 4.7 — Generally Available
Released April 16. Scores 64.3% on SWE-Bench Pro (~10 pp above Opus 4.6), resolves 3× more production tasks, and supports high-resolution images up to 2,576px / 3.75MP with 1:1 pixel-coordinate mapping — more than 3× the previous image resolution limit; also introduces a new xhigh effort level sitting between high and max, task budgets in beta (task-budgets-2026-03-13 header), improved filesystem-based memory, and automated cybersecurity safeguards (first GA Claude model to ship these). Pricing unchanged at $5 / $25 per MTok; available on the API, Amazon Bedrock, Google Vertex AI, Microsoft Foundry, and GitHub Copilot.
Opus 4.7 — Breaking API Changes
Three changes break existing Messages API code: (1) setting temperature, top_p, or top_k to any non-default value now returns a 400 error — remove these parameters; (2) thinking: {type: "enabled", budget_tokens: N} is removed — switch to thinking: {type: "adaptive"} with effort; (3) thinking content is omitted from responses by default (thinking.display defaults to "omitted") — set thinking.display: "summarized" to restore visible reasoning, or streaming products will show a long pause before output. Also: the new tokenizer may consume up to 35% more tokens than Opus 4.6 — update max_tokens and compaction triggers accordingly.
Claude Code v2.1.111
Released April 16. Ships Opus 4.7 with xhigh effort via an interactive /effort slider (arrow-key navigation); adds /ultrareview for comprehensive cloud-based parallel multi-agent code review; adds /less-permission-prompts skill that scans transcripts and auto-proposes a bash/MCP allowlist to cut permission prompts; begins progressive rollout of Windows PowerShell tool (CLAUDE_CODE_USE_POWERSHELL_TOOL opt-in/out); "Auto (match terminal)" theme option; plan files now named after their prompt (e.g., fix-auth-race-snug-otter.md). New keyboard shortcuts: Ctrl+U clears input buffer, Ctrl+Y restores deleted input, Ctrl+L forces full redraw. Fixes iTerm2+tmux display tearing, @ file-suggestion re-scanning in non-git directories, and spurious LSP diagnostics after edits.
Claude Code v2.1.112
Released April 16 (same-day hotfix). Fixes "claude-opus-4-7 is temporarily unavailable" error blocking auto mode immediately after Opus 4.7 launch.
Model Deprecations — Sonnet 4 and Opus 4
Announced April 14. claude-sonnet-4-20250514 and claude-opus-4-20250514 are deprecated; API retirement scheduled June 15, 2026. Recommended migration: Sonnet 4.6 for Sonnet 4, Opus 4.7 for Opus 4.
Vas Narasimhan Joins Long-Term Benefit Trust Board
Announced April 14. Roche CEO Vas Narasimhan appointed to Anthropic's Long-Term Benefit Trust board of directors.
🔌 Claude Code Plugins
None of the top 10 updated in window.
Unchanged in window: frontend-design, superpowers, context7, code-review, code-simplifier, github, feature-dev, playwright, skill-creator, claude-md-management
🛠️ Skills
None of the top 5 updated in window.
Unchanged in window: superpowers:brainstorming, superpowers:tdd, superpowers:systematic-debugging, superpowers:writing-plans, superpowers:executing-plans
🤖 Agents & Subagents
None of the top 5 updated in window.
Unchanged in window: superpowers:subagent-driven-development, claude-router:orchestrate, ralph-loop, claude-router:fast-executor, claude-mem:do
🔗 MCPs & Integrations
None of the top 5 updated in window.
Unchanged in window: github/github-mcp-server, microsoft/mcp (Azure + Fabric), modelcontextprotocol/servers, context7 mcp, claude-mem mcp-search
💡 Community — Workflows & Ideas
Claude Opus 4.7 — HN Discussion (Apr 16, 1,759 pts, 1,268 comments)
Developers surface two first-day concerns: adaptive thinking, when enabled, reportedly "chooses not to think when it should" — with community suggesting Anthropic's internal evals are measuring the wrong signal; and thinking content is now hidden by default, so streaming products that surfaced reasoning now show a silent pause before output unless display: "summarized" is explicitly configured.
📰 Quick Mentions
- AWS Blog (Apr 16) — Opus 4.7 on Amazon Bedrock; covers agentic task improvements and vision uplift for AWS customers.
- GitHub Changelog (Apr 16) — Claude Opus 4.7 available on GitHub Copilot with stronger multi-step and long-horizon task performance.
- Official Opus 4.7 migration guide — step-by-step checklist for all breaking changes; run
/claude-api migratein Claude Code to apply them automatically. - CNBC (Apr 16) — frames Opus 4.7 as Anthropic's "safer GA alternative" to the still-withheld Mythos Preview.
- anthropics/claude-plugins-official (Apr 16) — Supabase plugin re-pointed to community-maintained
supabase-community/supabase-pluginrepo. - Claude Code v2.1.109 (Apr 15) — minor: improved extended-thinking indicator with rotating progress hint during long operations.