SAN FRANCISCO — April 16, 2026 — Anthropic releases Claude Opus 4.7 on Thursday, its most capable commercially available AI model to date — arriving alongside a candid acknowledgement from the company that an even more powerful system, Claude Mythos Preview, remains locked away due to safety concerns. The release marks a meaningful upgrade from Opus 4.6, with particular improvements in advanced software engineering, vision capabilities, and a new architectural feature designed to give developers finer control over complex reasoning tasks.
Claude Opus 4.7 is now generally available across all Claude products and the Anthropic API, and is available through Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry at the same pricing as Opus 4.6 — $5 per million input tokens and $25 per million output tokens.
Benchmark Performance
On agentic coding benchmarks, Claude Opus 4.7 records 87.6% on SWE-bench Verified and 64.3% on SWE-bench Pro — meaningfully ahead of Claude Opus 4.6. On Terminal-Bench 2.0, the model scores 69.4%. In knowledge work, it reaches 64.4% on Finance Agent v1.1. On the graduate-level GPQA Diamond reasoning benchmark, it achieves 94.2%. – Anthropic Releases Claude Opus 4.7.
| Benchmark | Claude Opus 4.7 | Claude Opus 4.6 |
|---|---|---|
| SWE-bench Verified | 87.6% | Lower — specific figure not published |
| SWE-bench Pro | 64.3% | Lower — specific figure not published |
| Terminal-Bench 2.0 | 69.4% | Lower — specific figure not published |
| GPQA Diamond | 94.2% | Not directly comparable |
| Finance Agent v1.1 | 64.4% | Not directly comparable |
| Context window | 1 million tokens | 1 million tokens |
Claude Opus 4.7 key benchmark results, April 2026. Source: Anthropic.
According to VentureBeat, Opus 4.7 beats both OpenAI’s GPT-5.4 and Google’s Gemini 3.1 Pro on agentic coding, scaled tool-use, and financial analysis benchmarks. However, GPT-5.4 still outperforms Opus 4.7 on agentic search (89.3% vs 79.3%) and multilingual Q&A, reflecting the increasingly tight competition at the frontier rather than a clean sweep.
Key New Features
The most architecturally notable addition is a new xhigh effort level — positioned between the existing high and max settings — that gives users finer control over the trade-off between reasoning depth and latency on hard problems. Anthropic recommends starting with high or xhigh for coding and agentic use cases. The model also introduces experimental task budgets, a system giving developers more control over how Claude manages its reasoning on longer multi-step tasks. – Anthropic Releases Claude Opus 4.7.
Vision capabilities have been substantially upgraded: the model can see images in greater resolution, with Anthropic describing the improvement as substantially better vision at 3.3 times higher resolution than Opus 4.6. The model is also described as more tasteful and creative when completing professional tasks, producing higher-quality interfaces, slides, and documents.
Within Claude Code, the release adds a new /ultrareview command designed to simulate a senior human reviewer — flagging subtle design flaws and logic gaps rather than surface-level syntax errors. Additionally, auto mode has been extended to Max plan users, allowing Claude to make autonomous decisions without constant permission prompts during longer workflows.
“Opus 4.7 is a notable improvement on Opus 4.6 in advanced software engineering, with particular gains on the most difficult tasks. Users report being able to hand off their hardest coding work — the kind that previously needed close supervision — to Opus 4.7 with confidence.”— Anthropic, official release statement, April 16, 2026
The Mythos Shadow
Anthropic’s release was accompanied by an unusual disclosure: the company publicly acknowledged that Claude Opus 4.7 is less capable than Claude Mythos Preview, a significantly more advanced model that has only been made available to a select group of technology and cybersecurity companies through the recently announced Project Glasswing initiative.
In a chart accompanying its announcement, Anthropic showed that Opus 4.7 beats Opus 4.6, ChatGPT 5.4, and Google Gemini 3.1 Pro in a number of key benchmarks — but Opus 4.7 still falls short of its Mythos Preview model, which has only been released to a handpicked group of tech and cybersecurity companies.
Anthropic said it does not plan to make Claude Mythos Preview generally available, but the release of Opus 4.7 serves as the testbed for new automated cybersecurity safeguards that the company hopes will eventually allow a broader release of Mythos-class models. Opus 4.7 is the first model deployed with safeguards that automatically detect and block requests indicating prohibited or high-risk cybersecurity uses.
Cybersecurity Programme
Anthropic said it will use the new release to test guardrails designed to prevent its model being used for cybersecurity attacks. The company encouraged security professionals interested in using Opus 4.7 for legitimate cybersecurity purposes — such as vulnerability research, penetration testing, and red-teaming — to apply through a new Cyber Verification Program.
ContextClaude Opus 4.7 is available now across all Claude products, the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. Pricing remains $5 per million input tokens and $25 per million output tokens — identical to Opus 4.6. Developers access the model via the claude-opus-4-7 API string.
Enterprise Reception
Early-access feedback from enterprise customers has been strong. Hex, a data platform, reported that Opus 4.7 lifted coding benchmark resolution by 13% over Opus 4.6 across 93 tasks, including four tasks neither Opus 4.6 nor Sonnet 4.6 could solve. A financial technology platform cited in Anthropic’s release noted the model’s combination of speed and precision as potentially game-changing for development velocity.
However, enterprises should be aware of a significant operational consideration: Opus 4.7 uses an updated tokenizer that can increase input token counts by 1.0–1.35x, and its tendency to think harder at high effort levels results in higher output token consumption. For applications where prompts are fragile and margins are thin, a phased rollout with re-tuning is advisable.
The release arrives as Anthropic reports $30 billion in annualised revenue as of early April 2026 — a figure that has grown from $9 billion at the end of 2025 — and while the company is fielding investor offers at a potential $800 billion valuation.
Related Coverage
- Anthropic Attracts $800 Billion Valuation Offers as Revenue Hits $30 Billion
- Perplexity AI vs Claude: Which Is Better in 2026?
- Perplexity Model Council: Claude Opus Now Available
Frequently Asked Questions
What is Claude Opus 4.7?
Claude Opus 4.7 is Anthropic’s latest commercially available AI model, released on April 16, 2026. It is an upgrade to Claude Opus 4.6 with improvements in advanced software engineering, vision (3.3× higher resolution), multi-step reasoning, and professional task quality. It scores 87.6% on SWE-bench Verified and 94.2% on GPQA Diamond. Pricing is $5/$25 per million tokens — unchanged from Opus 4.6.
What is the xhigh effort level in Claude Opus 4.7?
xhigh is a new reasoning effort level between the existing high and max settings, giving developers finer control over the trade-off between reasoning depth and response latency on difficult problems. Anthropic recommends starting with high or xhigh for coding and agentic tasks when using Opus 4.7.
How does Claude Opus 4.7 compare to GPT-5.4?
Opus 4.7 outperforms GPT-5.4 on agentic coding, scaled tool-use, agentic computer use, and financial analysis benchmarks. GPT-5.4 leads Opus 4.7 on agentic search (89.3% vs 79.3%) and multilingual Q&A. Neither model achieves a clean sweep across all categories, reflecting the increasingly competitive frontier AI landscape in April 2026.
Is Claude Mythos Preview available?
No — Claude Mythos Preview is not generally available. It has been released only to a select group of technology and cybersecurity companies through Project Glasswing. Anthropic has stated it does not plan to make Mythos broadly available, and is using Claude Opus 4.7’s deployment to test automated cybersecurity safeguards before eventually working toward a broader release of Mythos-class models.
