Claude Mythos Is Coming Before May 6: The 'Code with Claude' Event Is the Signal — Everything We Know About 2026's Most Anticipated Release
Anthropic's 'Code with Claude' event on May 6 mirrors the Claude 4 launch format. Claude Mythos (Capybara) is expected before that date. Here's everything we know.
On March 26, an Anthropic leak revealed Claude Mythos: a model described as "by far the most powerful ever built" and "too dangerous for general public release." Today, a precise signal has emerged: Anthropic is hosting a "Code with Claude" event on May 6 — the exact same format used to launch the Claude 4 series in May 2025. Mythos is coming before May 6. Here's everything we know.
The "Code with Claude" Signal: Why May 6 Is a Date, Not a Coincidence
In May 2025, Anthropic hosted an event called "Code with Claude." Live sessions. Live demos. The result: the launch of the Claude 4 series — Claude 4 Sonnet, then Opus. That format had never been used for anything other than a major product launch.
May 2026: same title. Same format. "Live demos of new capabilities." The signal was spotted by @AiBattle_ on X as early as March 26.
Anthropic doesn't host these events for nothing. They host them to launch models. The question is no longer "is Mythos coming soon?" It's: "does Mythos drop before or during the event?"
Prediction markets — Polymarket, Metaculus — give a clear answer. Mythos is the most likely release of April/early May 2026. The inferred timeline: expanded early access in mid-April. Official announcement around May 6.
Mythos vs Opus: A New Category, Not an Upgrade
This point is critical. Claude Mythos is not Claude Opus 5. It's not "Opus improved." It's a structurally new category in the Claude lineup — just as Opus was when it replaced the old naming convention.
The quote from internal documents first reported by Fortune on March 26 is unambiguous: "Larger and smarter than our Opus models, which were, until now, our most powerful."
Three areas of outperformance are documented against Claude Opus 4.6:
- Cybersecurity — the most dramatic gap, described as "unprecedented"
- Software programming — significant coding gains
- Academic reasoning — improved benchmark performance
The internal codename — Capybara — confirms this isn't an incremental update. It's a new tier in Anthropic's taxonomy. To grasp the scope of the original leak, you have to read what the documents said between the lines: Anthropic built something it doesn't yet know how to safely distribute.
| Model | Category | Use Case | Status |
|---|---|---|---|
| Claude Haiku | Fast / light | Volume, mobile | ✅ Available |
| Claude Sonnet | Balanced | Daily use, chat | ✅ Available |
| Claude Opus 4.6 | Powerful | Advanced reasoning | ✅ Available |
| Claude Mythos | New category | Cyber, code, research | 🔜 Before May 6 |
"Unprecedented in Cybersecurity" — Why Anthropic Fears Its Own Model
Here's the central paradox. Anthropic built the most powerful AI model in its history. And it's refusing to distribute it widely. The reason: it's too capable at offensive cybersecurity.
The precedents explain this caution.
Precedent 1: during an internal security test, Claude Opus was turned into a "malware factory" in 8 hours. A researcher managed to get it to generate functional attack tools with zero prior cybersecurity knowledge.
Precedent 2: a Chinese state-sponsored campaign using Claude Code infiltrated 30 organizations before Anthropic shut it down. The attack used the autonomous agent mode to pivot between compromised systems.
If Opus already enabled that, Mythos is in another category entirely. Anthropic knows it. Hence the asymmetric deployment strategy: give access to cyber defenders first. The goal: level the playing field before potential attackers gain access to the model.
As noted in the Futurism full analysis, this isn't crisis communication. It's a deliberate layered deployment strategy.
Why Anthropic Can't Wait Until Q3
Some outlets misread "03 | 2026" in the internal documents as Q3 2026. It was the document date — March 2026. A source close to Anthropic confirms: the release is planned well before Q3.
The competitive pressure explains why.
Gemini 2.5 Pro (Google) has held the #1 spot on LMArena — the reference leaderboard for AI models — since mid-March 2026. Gemma 4 (also Google) launched on April 1 with an AIME score of 89% in open-source under Apache 2.0. OpenAI Spud — OpenAI's next flagship model — has completed pre-training and entered fine-tuning. Expected release: before end of May. Grok 4 (Elon Musk's xAI) is announced for the coming weeks.
| Model | Lab | Status | Timeline |
|---|---|---|---|
| Gemini 2.5 Pro | ✅ #1 LMArena | Available | |
| Gemma 4 31B | ✅ Open-source | Available | |
| Claude Mythos | Anthropic | 🔜 Imminent | Before May 6 |
| OpenAI Spud | OpenAI | 🔜 Fine-tuning | Before end of May |
| Grok 4 | xAI | 🔜 Expected | April-May |
If Anthropic waits until summer, it arrives third on its own turf: reasoning, safety, code. Releasing Mythos in April-May means releasing alongside Spud. It's the direct showdown for the title of "best frontier model of 2026."
Dario Amodei has played the "safety first" card since Anthropic's founding. But the industry doesn't wait.
Restricted Early Access: Who Gets Mythos First?
According to the leaked internal documents, deployment follows a precise order.
Priority 1 — cyber defenders. Organizations actively strengthening their IT security systems. The goal: give them a defensive advantage before the model becomes accessible to attackers.
Priority 2 — an invite-only CEO summit in Europe. Dario Amodei was scheduled to present Mythos to European CEOs at a private event in the United Kingdom.
Priority 3 — existing Anthropic enterprise customers. Those already paying for API access to Claude models.
General public: limited access in the short term. Likely through claude.ai with strict rate limiting — a system that caps the number of requests per user — and premium pricing. Mythos's high operational cost is cited as the second reason for the restricted rollout, after the cyber risk.
Bottom line: if you're using claude.ai's free tier, you may not see Mythos before summer. If you're an enterprise customer, likely mid-April.
Key Takeaways
- Claude Mythos (codename Capybara) is Anthropic's next flagship — a category above Opus, described as "by far the most powerful ever built"
- Release signal: "Code with Claude" event on May 6, 2026 with "live demos of new capabilities" — the exact same format used to launch the Claude 4 series in May 2025
- Likely timeline: enterprise early access mid-April, official announcement around May 6, 2026
- Three areas of outperformance vs Claude Opus 4.6: cybersecurity ("unprecedented" gap), code, reasoning
- Asymmetric deployment: cyber defenders first, general public with limited access + premium pricing in the short term
The March 26 leak was accidental. The May 6 event is intentional. Between the two: Anthropic built the most powerful model in its history, got scared of its own creation, suffered two major leaks in one week, sued the Pentagon, and acquired a biotech startup for $400 million. Mythos arrives in this context — not as a quiet launch, but as Anthropic's answer to the most chaotic week in its existence. "Code with Claude." May 6. Mark the date.
