Claude Opus 4.7 Launches With 13% Coding Gains and New Claude Code Tools
Anthropic ships Claude Opus 4.7 today with notable gains on agentic coding benchmarks, 3x better image resolution, a new xhigh effort level, and a /ultrareview command in Claude Code — while openly conceding the model still trails the unreleased Mythos Preview.
Image by Anthropic
Claude Opus 4.7 Launches With Meaningful Coding Gains, Sharper Vision, and New Claude Code Tooling
Anthropic today released Claude Opus 4.7, the company's most capable generally available model. The release arrives amid weeks of developer frustration with Opus 4.6 — and delivers a measurable improvement across agentic coding, vision, and long-horizon task execution — while openly positioning itself as a stepping stone toward the still-unreleased Claude Mythos Preview.
The announcement was made via Anthropic's official blog and X account on April 16, 2026, with broad coverage from CNBC, Axios, and 9to5Mac confirming the details. Cursor, Replit, Notion, Bolt, and Devin all provided early-access benchmarks alongside the launch.
What Changed in Opus 4.7
The headline improvement is coding performance. On Cursor's internal CursorBench, Opus 4.7 scored 70% versus Opus 4.6 at 58% — a 12-point jump. On a 93-task coding benchmark run by GitHub Next, Opus 4.7 lifted task resolution by 13% over Opus 4.6, including four problems that neither Opus 4.6 nor Sonnet 4.6 could solve at all. On Rakuten's SWE-Bench variant, Opus 4.7 resolves 3x more production tasks than its predecessor.
The model introduces a new ability to self-check its own outputs before reporting back — what Anthropic calls "devising ways to verify its own outputs." In testing by Hex, low-effort Opus 4.7 was roughly equivalent to medium-effort Opus 4.6, meaning the same quality at lower compute cost.
Vision is a significant upgrade: Opus 4.7 now accepts images up to 2,576 pixels on the long edge (approximately 3.75 megapixels), more than three times the resolution ceiling of prior Claude models. For computer-use agents, dense screenshot reading, or technical diagram extraction, this is a practical change. XBOW, an autonomous penetration testing firm, reported their visual-acuity benchmark scores jumped from 54.5% on Opus 4.6 to 98.5% on Opus 4.7.
New Claude Code Features
The release bundles several Claude Code updates developers should know about:
/ultrareviewcommand: A new slash command that runs a dedicated review session across your changes, flagging what a careful code reviewer would catch — bugs, design issues, edge cases. Pro and Max Claude Code users get three free ultrareviews to try it.xhigheffort level: A new effort tier sits betweenhighandmax, giving developers finer control over the reasoning-vs-latency tradeoff. Claude Code's default effort has been raised toxhighfor all plans. Anthropic recommends starting athighorxhighfor coding and agentic use cases.- Auto mode for Max users: Auto mode — which lets Claude make permission decisions on your behalf so longer tasks run with fewer interruptions — has been extended to Max plan subscribers.
- Task budgets (API, public beta): API users can now set token budgets on longer runs, giving developers more control over how Claude allocates its reasoning effort across multi-step tasks.
The Mythos Context
Anthropic was unusually candid about where Opus 4.7 sits in their model hierarchy. The company publicly acknowledged that the new model does not match Claude Mythos Preview — their most powerful model, which remains in a limited, invitation-only release through Project Glasswing, announced last week.
In a benchmark chart accompanying the announcement, Opus 4.7 beats Opus 4.6, GPT-5.4, and Gemini 3.1 Pro across a range of evaluations, but trails Mythos Preview. Anthropic framed Opus 4.7 as the first model to receive experimental cyber safeguards that will eventually allow a broader release of Mythos-class capability. Those safeguards automatically detect and block requests for prohibited or high-risk cybersecurity uses. Security professionals doing legitimate vulnerability research, penetration testing, or red-teaming can apply through Anthropic's new Cyber Verification Program.
This framing addresses a secondary thread in the release: the wave of developer complaints in recent weeks that Opus 4.6 had quietly regressed. An AMD senior director's GitHub post — "Claude has regressed to the point it cannot be trusted to perform complex engineering" — circulated widely and put Anthropic on the defensive. Anthropic denied redirecting compute from Opus to Mythos.
Pricing and Availability
Pricing is unchanged from Opus 4.6: $5 per million input tokens and $25 per million output tokens, with up to 90% savings via prompt caching and 50% via batch processing. Opus 4.7 is available today across all Claude products (Pro, Max, Team, Enterprise), the Claude API (claude-opus-4-7), Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
Developers migrating from Opus 4.6 should note two token-usage changes: Opus 4.7 uses an updated tokenizer that can produce 1.0–1.35x more tokens for the same input depending on content type, and the model thinks more at higher effort levels, producing more output tokens on hard problems. Anthropic has published a migration guide at the Claude Platform docs.
What's Unconfirmed
Anthropic has not disclosed Opus 4.7's full training details, architecture changes, or context window beyond the existing 200K-token limit in the consumer product (1M via API for enterprise). The exact timeline for a broader Mythos Preview release — and whether Opus 4.7's cyber safeguard testing will accelerate or delay it — remains open. No information has been shared on when a Sonnet 4.7 or Haiku update will follow.
