On April 7, 2026, Anthropic quietly published a technical report on a model called Claude Mythos — and it immediately became clear why they did it without a big announcement. The model leads in 17 out of 18 benchmarks among all existing AI systems. But only a select few will be able to access it.
What is Claude Mythos and Why It Matters
Claude Mythos is the new flagship model from Anthropic, creators of Claude AI. According to Anthropic’s own testing, it outperforms all competitors in almost every metric.
The only test where Mythos didn’t take first place was MMMLU (language understanding and knowledge). There, Mythos scored 92.7 points, matching the range of Gemini 3.1 Pro (92.6–93.6). Essentially a tie.
Comparison with Current Flagships:
| Model | Overall Rating | Leads In |
|---|---|---|
| Claude Mythos | 17/18 benchmarks | Cybersecurity, coding, analysis |
| Claude Opus 4.6 | Top 3 | Coding (SWE-bench 80.8%) |
| GPT-5.4 | Top 2 | Overall rating |
| Gemini 3.1 Pro | Top 2 | MMMLU |
Why Anthropic Isn’t Releasing Mythos Publicly
And here’s where it gets interesting.
Anthropic doesn’t plan a public release of Claude Mythos. Instead, the company launched a closed program called Project Glasswing — access to the model is granted only to partners with credits worth at least $100 million.
H3: Official Reason — Cybersecurity
Mythos showed exceptional results specifically in cybersecurity: Cybench tests, CyberGym, and a specialized Firefox 147 exploitation evaluation. The model is so powerful at identifying vulnerabilities that Anthropic decided not to risk public access.
H3: Real Reason — GPU Shortage
Analysts point to another reason: lack of computing power. The facts speak for themselves:
- OpenAI shut down Sora because it was burning ~$1 million per day with only $2.1 million in total revenue
- Anthropic disconnected Claude subscribers from third-party agent systems like OpenClaw
- In the first eight days of April 2026, five incidents with Claude Opus and Sonnet availability were recorded
The conclusion is obvious: leading AI labs are rationing GPUs and directing them to the most profitable corporate clients.
Anthropic vs OpenAI — An Accelerating Race
Against the backdrop of Mythos’s release, Anthropic’s financial performance looks impressive.
Anthropic in Numbers (early 2026):
- Annual revenue: ~$19 billion — 10x growth in a year
- Claude Code separately: $2.5 billion annual run rate
- Corporate API market share: 32% (vs. 25% for OpenAI)
- Company valuation: $350 billion after the latest round
According to Epoch AI forecasts, if both players maintain their current pace — Anthropic could overtake OpenAI in revenue as early as 2026–2027.
Claude Managed Agents — A New Product for Business
Alongside Mythos, Anthropic announced Claude Managed Agents — a product for corporate clients who want to build AI agents without deep technical knowledge.
H3: What This Gives Businesses
Typically, deploying an autonomous AI agent in a corporate environment requires months of engineering work. Claude Managed Agents solves this out of the box:
- Built-in agent memory
- Permissions and restrictions system
- Isolated execution environment (sandboxing)
- Ready-made agent templates
The goal is to allow companies to skip most of the infrastructure work and immediately build useful products based on Claude.
What We Know About Claude 5 and Future Plans
Mythos isn’t the final chapter. According to industry sources, Anthropic is actively working on the next generation of models.
Claude Code — a tool for programmers that integrates directly into code editors — is already generating $2.5 billion in annual revenue and growing more than 70% monthly. This signals where Anthropic is directing its main efforts: enterprise coding and automation.
What This Means for Ukrainian Claude Users
For most Ukrainians using Claude AI through claude.ai — nothing changes. Claude Opus 4.6 and Sonnet 4.6 remain available.
Mythos is a corporate tool for the biggest market players. But its existence means one thing: Anthropic is seriously ahead of competitors in technical capabilities right now.
And that’s good for all of us: the competition between Anthropic and OpenAI is the main driver of why AI tools are getting better every month.
Material prepared by the TechVisor team — practical IT media for people. Sources: R&D World Online, TechCrunch, Anthropic system card (04/07–09/2026)
