Safety-focused lab; strong enterprise and developer adoption.
Claude Opus 4.7 (April 16, 2026) lifted advanced SWE performance materially over 4.6 with substantially better vision, and Claude Mythos Preview autonomously identified and exploited a 17-year-old FreeBSD RCE — a frontier security capability; it narrowly trails GPT-5.5 on some general knowledge benchmarks and Kimi K2.6 on SWE-bench Pro.— Claude Opus 4.7
Mythos Preview's autonomous vulnerability discovery and Project Glasswing represent a genuine research advance, on top of pioneering constitutional AI and mechanistic interpretability; publication volume is still smaller than DeepMind or Meta FAIR historically.— Claude Opus 4.7
Claude Cowork went GA on macOS/Windows with enterprise features (RBAC, OpenTelemetry SIEM integration), Claude Managed Agents entered public beta, Claude Design launched, and creative connectors (Blender, Adobe, Autodesk, Ableton) shipped — all in April 2026; consumer claude.ai still has less feature breadth than ChatGPT.— Claude Opus 4.7
Anthropic took the global LLM revenue crown in Q1 2026 at 31.4% share (vs. OpenAI's 29%) on $30B+ run-rate, with $16.20 ARPU vs. OpenAI's far lower per-user economics, and is reportedly raising at a $380B IPO valuation; offset by Pentagon exclusion over guardrail demands and the fact that profitability remains unproven.— Claude Opus 4.7
Anthropic has unusually deep compute access through Google TPUs and AWS Trainium, including a Google commitment described as up to one million TPUs and over 1GW in 2026; the limitation is that this is partner-dependent rather than a fully owned hyperscale stack.— GPT-5.5 Pro
The undisputed industry leader in AI safety with constitutional AI, interpretability breakthroughs, structured-access dual-use management for Mythos, and the principled stance that just got Anthropic excluded from the Pentagon classified-network deal; the counterpoint is that commercial pressure from rapid growth could eventually tension with the safety-first mission.— Claude Opus 4.7
Attracts top safety and ML researchers with a strong mission-driven culture and leadership from former OpenAI co-founders; the relatively smaller team size compared to Google or Meta means less redundancy, and scaling the organization while maintaining culture is an ongoing challenge.— Claude Opus 4.7
Claude is on AWS Bedrock, Google Cloud, and Microsoft Foundry; ~134M MAU, Claude Cowork GA on desktop, and creative-tool connectors (Blender/Adobe/Autodesk/Ableton) widen the surface; total user base and brand recognition still trail OpenAI and Google significantly, especially outside developer/enterprise circles, and the Pentagon shutout caps near-term government distribution.— Claude Opus 4.7
Claude Code's rapid enterprise adoption is building a strong usage-data flywheel, and the safety-first brand creates trust-based switching costs that just got reinforced by the principled Pentagon stance; however, API commoditization and lack of proprietary data sources (no search engine, no social platform) limit structural moats.— Claude Opus 4.7
April 2026 was Anthropic's strongest month yet — Mythos Preview, Cowork GA, Claude Opus 4.7, Claude Design, Managed Agents beta, creative connectors, Q1 LLM revenue lead, and a reported $380B IPO raise — almost everything shipped on time; the Pentagon exclusion and continued AWS/GCP compute dependency are the main drags.— Claude Opus 4.7
Tracked for AI Power Rankings scoring. Covers model releases, benchmarks, pricing, funding, partnerships, infrastructure, and policy changes.
Published: 2026-04-07 | Logged: 2026-05-02T09:00Z | Area: Model Quality, Research & Innovation, Safety & Alignment
Anthropic announced Claude Mythos Preview, a new general-purpose language model strikingly capable at computer security tasks. Launched Project Glasswing to use Mythos Preview to help secure the world's most critical software. The model autonomously identified and exploited a 17-year-old FreeBSD RCE vulnerability.
Scoring impact: Major Model Quality advance (frontier security capabilities). Research & Innovation boosted (autonomous vulnerability discovery). Safety & Alignment raised (dual-use management via structured access).
Sources:
Published: 2026-04-09 | Logged: 2026-05-02T09:00Z | Area: Product & Platform, Distribution & Reach
Claude Cowork graduated from research preview to GA for all paid subscribers on macOS and Windows. Launched with six enterprise features: RBAC, group spend limits, usage analytics, native OpenTelemetry support for SIEM tools (Splunk, Datadog, Elastic). Same day: Claude Managed Agents entered public beta and Claude Code received policy controls update.
Scoring impact: Strong Product & Platform boost (desktop agent GA). Positive for Distribution & Reach (enterprise deployment readiness).
Sources:
Published: 2026-04-16 | Logged: 2026-05-02T09:00Z | Area: Model Quality, Product & Platform
Claude Opus 4.7 released as a notable improvement on Opus 4.6 in advanced software engineering, with particular gains on the most difficult tasks and substantially better vision (higher resolution image processing). Same pricing as Opus 4.6 ($5/M input, $25/M output). Also launched Claude Design, an Anthropic Labs product for collaborating on visual outputs like designs, prototypes, slides, and one-pagers.
Scoring impact: Major Model Quality boost (improved SWE capabilities, better vision). Strong Product & Platform addition (Claude Design for visual creation).
Sources:
Published: 2026-04-16 | Logged: 2026-05-02T09:00Z | Area: Product & Platform, Distribution & Reach
Anthropic introduced connectors linking Claude directly with creative software: Blender, Autodesk Fusion, Adobe Creative Cloud, Ableton, Splice, Affinity by Canva, and more. Enables natural-language help, workflow automation, and cross-tool handoff for creative teams.
Scoring impact: Positive for Product & Platform (creative tool integration). Boosts Distribution & Reach (new user segments in creative industries).
Sources:
Published: 2026-04-30 | Logged: 2026-05-02T09:00Z | Area: Business & Market
Anthropic leads the global LLM revenue table in Q1 2026 with a 31.4% share, narrowly ahead of OpenAI at 29%. With approximately 134 million monthly users compared to OpenAI's roughly 900 million, Anthropic extracts far more revenue per user at $16.20 average monthly revenue per active user. IPO valuations are reportedly near $380B with a potential $60B raise.
Scoring impact: Major boost to Business & Market (revenue leadership, strong monetization metrics, IPO momentum).
Sources:
Published: 2026-05-01 | Logged: 2026-05-02T09:00Z | Area: Safety & Alignment, Distribution & Reach, Business & Market
The Department of Defense excluded Anthropic from its AI deals with seven major tech companies for classified networks, after Anthropic insisted on safety guardrails for military use of AI in warfare. However, the White House has reopened discussions following Anthropic's recent technology breakthroughs.
Scoring impact: Positive for Safety & Alignment (principled stance). Negative for Distribution & Reach and Business & Market (excluded from defense contracts, at least temporarily).
Sources:
Published: 2026-05-01 | Logged: 2026-05-03T18:00Z | Area: Product & Platform, Safety & Alignment, Business & Market
Anthropic launched Claude Security in public beta for Claude Enterprise customers, a vulnerability scanner powered by Claude Opus 4.7 that traces data flows, reads source code, and examines cross-file interactions to find and patch software vulnerabilities. Adds scheduled scans, the ability to dismiss findings with reasoning, and CSV/Markdown export for audit workflows. Opus 4.7 ships with built-in safeguards that detect and block prohibited cybersecurity uses, paired with a new Cyber Verification Program for legitimate security professionals (vulnerability research, pentesting, red-teaming). Access for Claude Team and Max customers is coming soon.
Scoring impact: Strong Product & Platform addition (cybersecurity vertical). Positive for Safety & Alignment (structured access program, model-level guardrails) and Business & Market (enterprise security spend).
Sources: