Catching Up on Everything Claude — From Opus 4.8 to Fable 5, Three Months That Changed the Landscape

[Editor’s note, July 25, 2026] After this article was finalized, Anthropic released Claude Opus 5 on July 24 — same $5/$25 API pricing as Opus 4.8, 1M context, 128K output, and the new default model on the Max plan. The facts below are accurate as of mid-July 2026; factor Opus 5 into any model-selection decisions. A full review is coming soon.

In late April 2026, we published a complete roadmap of Claude’s spring lineup. Back then the flagship was Claude Opus 4.7, and the conversation revolved around auto mode and xhigh effort. Just three months later, the flagship has changed hands, a new tier called the Mythos class now sits on top of the model hierarchy, and the top model was even taken offline for weeks by a US export restriction — an event without precedent in the AI industry.

If you caught fragments of the news but never connected them into one picture, this article is for you. It ties together everything that happened to Claude from late May through mid-July 2026, in chronological order. By the end, you should know what is available now, what changed, and where you should start. Every date, price, and spec here has been checked against Anthropic’s official announcements and documentation; anything based only on press reports is flagged as such.

The “common sense” of three months ago no longer applies

Recall the Claude environment as of spring. The flagship was Opus 4.7, xhigh effort was the recommended setting for deep reasoning, and Claude Code had just gained auto mode. Many readers built their workflows on that picture. Starting May 28, that picture began to fall apart — and this time the change was not just incremental performance. The model hierarchy itself was restructured, and the supply of AI became a geopolitical variable. Anthropic now moves three layers — models, tools, and delivery terms — in parallel, week after week.

The six weeks in one table

Notice the density: a flagship generation change, a brand-new tier, and a government-ordered suspension and restart — a year’s worth of change compressed into six weeks. Let’s walk through each piece.

Claude Opus 4.8 — the flagship that handed you the thinking dial

Released on May 28, 2026, Claude Opus 4.8 kept API pricing at $5 per million input tokens and $25 per million output tokens, unchanged from Opus 4.7. What expanded was freedom of use. The headline change is the generalization of effort control: on every claude.ai and Cowork plan, you now choose how much compute goes into each response. The default is high — roughly Opus 4.7’s default token consumption but with better quality — and you can step up to extra or max to raise the hit rate on genuinely hard problems.

The other highlight is fast mode: about 2.5x standard output speed at $10/$50, three times cheaper than fast mode on previous models. It is built for work where waiting itself is the cost, such as interactive coding sessions. Claude Code also gained dynamic workflows as a research preview — Claude writes its own orchestration script and coordinates hundreds of parallel subagents in a single session.

Claude Fable 5 — the Mythos class above Opus

On June 9, Anthropic announced Claude Fable 5 and Claude Mythos 5, adding a new capability tier — the Mythos class — above the familiar Haiku, Sonnet, and Opus lineup. API pricing is $10/$50, exactly double Opus 4.8. Anthropic positions Fable 5 as the most capable model it has ever made generally available: state of the art on nearly every capability benchmark tested, and the top score among frontier models on Cognition’s FrontierCode — achieved even at medium effort.

The safety design deserves attention. Fable 5 ships with safety classifiers covering three categories — cyberattack capability, biology and chemistry, and model-capability extraction (distillation). Requests flagged by the classifiers are not answered by Fable 5 itself; they fall back to Opus 4.8. On top of that, all Mythos-class traffic is retained for 30 days. On subscriptions, Fable 5 was included at no extra cost from the announcement through June 22, moving to usage credits on June 23 — a plan that the suspension soon disrupted.

Claude Mythos 5 — same model, a different tap

Mythos 5 is the same underlying model as Fable 5 with the same specs and pricing. The difference is the plumbing: it has no safety classifiers and is not generally available. It is delivered through a vetting framework to approved organizations working on cyber defense and critical infrastructure, plus screened biology researchers. Same water source, two taps — one filtered for everyone, one unfiltered for licensed operators.

The suspension and the comeback — June 12 to July 1

Just three days after the announcement, on Friday June 12, the US government applied export controls to Fable 5 and Mythos 5 and both models were suspended. Press reports cited national-security concerns, with researchers’ jailbreak findings in the background — while some experts argued the risk was overstated (Al Jazeera). The restrictions were lifted on June 30, and on July 1 Fable 5 was redeployed globally across the Claude Platform, Claude.ai, Claude Code, and Cowork, with cloud availability through AWS, Google Cloud, and Microsoft Foundry restored progressively.

For the relaunch, Anthropic says it deployed improved classifiers that block the reported jailbreak technique in over 99% of cases. Paid plans could use Fable 5 for up to 50% of weekly usage limits through July 7, then moved to usage credits. The lesson is clear: for top-tier models, supply stability is now part of the spec sheet. A workflow built on a single model is a data center with one power feed.

Claude Sonnet 5 — a quiet change of the default

While the headlines went to the Mythos class, Claude Sonnet 5 quietly became the default model on Pro, Team Standard, and Enterprise seats between late June and early July. Today’s Sonnet 5 carries a 1M-token context window and adaptive thinking as standard. Much of what used to require the flagship — whole-repository comprehension, long spec cross-checks, multi-document reviews — now fits inside the default model. The trigger for calling a flagship has shifted from “how long is the context” to “how hard is the reasoning.”

Claude Code ships weekly — summer highlights

fallbackModel (week of June 8): configure up to three fallback models tried in order — the tool-side answer to supply risk. Nested subagents (same week): subagents can now spawn their own subagents, with background chains capped at five levels. /cd and –safe-mode: move the working directory mid-conversation without losing the prompt cache, and boot with all customization disabled for troubleshooting.

claude mcp login (week of June 22): authenticate MCP servers straight from the shell. Artifacts (mid-June onward): publish session outputs as shared pages on claude.ai — beta for Team and Enterprise, extended in mid-July with public links and viewer-side MCP connectors. Desktop built-in browser and /doctor (early July): Claude can open and operate web pages itself, and self-diagnose setup issues. /fork and screen-reader mode (mid-July): duplicate a conversation for parallel exploration, and a linear-text mode for assistive tech. auto mode expanded to Amazon Bedrock, Google Cloud, and Microsoft Foundry.

On claude.ai, inline Artifact editing arrived June 12 — highlight a passage, write an instruction, and only that part is rewritten. On July 14, Enterprise gained entitlement management for models and effort levels, letting administrators control who can use which model at which depth. Managing an organization’s “intelligence budget” is now a real design task.

What to change today — guidance by role

Developers: start by revisiting your Claude Code model configuration. Insure against supply risk with fallbackModel, run daily tasks on Sonnet 5 with its 1M context, handle hard problems with Opus 4.8 and effort tuning, and spend Fable 5 credits only on the genuinely tough ones. Reserve fast mode for interactive work where waiting breaks your rhythm.

Technical decision-makers: treat the June suspension as a prompt to audit where AI sits in your business. If any workflow assumes a single model, put an abstraction layer and fallback targets on the agenda.

Makers: code-CAD workflows benefit most directly from this generation — longer code context means fewer revision round-trips in parametric design, while your prompting patterns carry over unchanged.

Investors and industry watchers: June was the moment AI stepped into regulated-industry territory. Frontier model supply is now a function of security policy as much as technical roadmaps. Add “continuity of supply” and “government relations” to your evaluation checklist.

Primary sources

Introducing Claude Opus 4.8; Claude Fable 5 and Claude Mythos 5; Redeploying Claude Fable 5 (Anthropic official announcements); Claude Code What’s New (official documentation); Claude Release Notes (official help center); US lifts restrictions on Anthropic AI models (Al Jazeera).

ブラウザだけでできる本格的なAI画像生成【ConoHa AI Canvas】

swiftwand

X

AIを使って、毎日の生活をもっと快適にするアイデアや将来像を発信しています。 初心者にもわかりやすく、すぐに取り入れられる実践的な情報をお届けします。 Sharing ideas and visions for a better daily life with AI. Practical tips that anyone can start using right away.