↓ Skip to main content
sdd.sh
sdd.sh
  1. Articles/

Code with Claude Tokyo Is 9 Days Away — and the Signals Point to Mythos

·1245 words·6 mins·
Author
Florent Clairambault
CTO & Software engineer
Table of Contents

Nine days from now, Anthropic holds its third developer conference of 2026.

Code with Claude Tokyo lands June 10 at a venue in Shibuya, with a second day added June 11 after registrations exceeded capacity — indie developers and early-stage founders only on the extension day. The main event runs three parallel tracks: Research (model capabilities and direction), Claude Platform (production agent deployment), and Claude Code (scaling autonomous workflows across real engineering organizations). Live simultaneous interpretation runs in both directions — Japanese and English — throughout.

This is worth paying attention to, and not just if you’re in Tokyo.

The Code with Claude Pattern
#

Anthropic runs these events differently than most vendor conferences. Code with Claude San Francisco in May was not a marketing showcase. It was a product launch platform. In a single day, Anthropic announced: 300MW of compute from SpaceX’s Colossus facility doubling Claude Code rate limits, Code Review GA at $15–25 per PR with multi-agent reviewers, Managed Agents Dreaming and Outcomes in public beta, and the Mercado Libre case study (targeting 90% autonomous coding across 23,000 engineers by Q3 2026).

That’s four significant product moves in one event. London, held the same week as Google I/O, dropped quieter announcements but confirmed the global enterprise push was real.

Tokyo is the third stop. If the pattern holds, it is not a celebration of what shipped — it’s where the next thing drops.

The Mythos Signal
#

In April, Anthropic confirmed what the leaked Project Glasswing documents had suggested: Mythos Preview is the most capable model Anthropic has built, restricted to nine enterprise security partners for defensive use, and described internally as “too dangerous to release broadly.” The 2026 USA Mathematical Olympiad score (97.6%, vs. Opus 4.6’s 42.3%) made the capability gap visceral. The UK AI Security Institute’s 73% success rate on expert-level penetration tasks — tasks that no model could complete at all in April 2025 — made the danger framing credible.

For most of April and May, “Mythos release” was a question of when safeguards would catch up to capability, not a near-term calendar item.

Two things shifted in the last two weeks.

On May 25, The Register published a piece titled “Anthropic to release Mythos-class models to the public,” citing internal sources confirming Anthropic had moved from “the model is too dangerous” to “we intend to release it publicly once safeguards are adequate.” That framing shift matters. “Too dangerous to release” is a permanent status. “Releasing once safeguards are adequate” is a roadmap commitment.

Around the same time, TestingCatalog published documentation suggesting Anthropic is specifically preparing “Mythos 1” for deployment on Claude Code and Claude Security. Not Mythos Preview — a versioned release candidate. The pricing from the Glasswing preview ($25/$125 per million tokens) is consistent with a flagship model positioned above Opus 4.8.

Neither source is an official Anthropic announcement. But Anthropic rarely announces anything before the day it ships. The company announced Claude Opus 4.8 with zero pre-announcement. Code Review GA was announced the morning of Code with Claude SF. The pattern is: you find out when it drops.

What Mythos 1 on Claude Code Would Actually Mean
#

Set aside the security benchmarks for a moment. The coding capability gap is what matters for most developers.

The 2026 USA Mathematical Olympiad result is a proxy for reasoning depth in complex problem decomposition — the kind of multi-step, constraint-laden thinking that separates a model that can write a function from one that can architect a system. A 55-percentage-point gap over Opus 4.6 is not a benchmark rounding error. It’s a different tier of capability.

For Claude Code specifically: Anthropic’s Dynamic Workflows feature (shipped with Opus 4.8) already enables hundreds of parallel subagents coordinating on codebase-scale tasks. Mythos 1 as the reasoning engine for that orchestration layer — the model that decides how to decompose a task, which agents to spawn, and how to validate their output — would represent a qualitative change in what autonomous coding can accomplish.

The comparison point is Opus 4.8’s 69.2% SWE-bench Pro score. If Mythos 1 carries a similar gap over Opus 4.8 as it does over Opus 4.6, you’re looking at a coding model that handles problems the current generation can’t reliably attempt.

What Else Tokyo Could Bring
#

Sonnet 4.8 is the other active leak. The NPM source map disclosure from March confirmed the model exists; subsequent leaks suggest vision capabilities at ~98% and an additional 12-point coding benchmark lift over Sonnet 4.6. Historically, Anthropic releases Opus and Sonnet generations within weeks of each other. Opus 4.8 shipped May 28. Mid-June is the historical window for Sonnet 4.8. Code with Claude Tokyo is June 10.

Beyond model releases, the three-track structure of the event tells you what Anthropic thinks its developer audience cares about: production deployment (Claude Platform track) and real-world Claude Code adoption at scale (Claude Code track). The “indie developers and early-stage founders” framing for the June 11 extension day is notable — Anthropic is explicitly courting the long tail, not just enterprise.

Given the recent Pro plan billing controversy (Anthropic briefly tested removing Claude Code from the $20 Pro tier in April), the indie developer focus reads partly as a trust-repair gesture, partly as genuine expansion strategy. The developer market outside Fortune 500 is where usage grows fastest.

Why Tokyo Specifically
#

Japan is not a symbolic stop. Anthropic’s enterprise presence in Asia-Pacific is growing, and the Japanese developer market is sizable and technically sophisticated. The Tokyo event runs sessions in both languages, and the agenda explicitly marks which sessions are Japanese-language — suggesting Anthropic is not just shipping a translated version of the SF event but building local content.

The April story about Apple’s 200 Siri engineers attending AI coding bootcamps — with Apple teams already running Claude Code and the Siri team identified as a “laggard inside Apple” — is also Tokyo-relevant. WWDC is June 8, two days before Code with Claude Tokyo. Apple is expected to announce the Siri Extensions framework that puts Claude as a first-class AI provider on every iPhone. The temporal proximity of WWDC and Code with Claude Tokyo is not likely accidental.

If Anthropic has news about Siri Extensions and Claude Code integration, or about the iOS 27 developer APIs for building Claude-backed experiences, Tokyo is the right place to tell that story to the developer audience most likely to build with it.

What to Watch
#

If you’re tracking the AI coding tools space, the next ten days matter.

June 8 (WWDC): Apple announces iOS 27, Core AI, and the Siri Extensions framework. Claude confirmed as a launch partner. Watch for any developer API details.

June 10 (Code with Claude Tokyo): Product keynote. Three tracks. The Research track is where model announcements land. If Mythos 1 or Sonnet 4.8 ships, it comes with the keynote.

June 11 (Tokyo Extended): Indie developer and early-stage founder day. Often where Anthropic previews what’s coming to lower-tier plans.

The SF event in May was the biggest single-day Anthropic announcement since the original Claude Code launch. Tokyo has the same infrastructure in place. The only question is what’s ready.

Sources:

Related