0

10 things Claude Code can do today that it couldn't yesterday

Anthropic shipped Fable 5 — its first Mythos-class model — on June 9. Here's what it (and the latest Claude Code) unlocks that genuinely wasn't possible the day before.

By YoloBytes Team ·

Yesterday — June 9, 2026 — Anthropic released Claude Fable 5, its first Mythos-class model made generally available. Mythos sits a tier above the Opus class, and Fable 5 is the safeguarded version the public can use today (its sibling, Mythos 5, keeps some safeguards lifted for approved customers). Around the same window, Claude Code shipped features that lean right into it.

So, in the spirit of "what can it do that it couldn't yesterday" — here are ten, each pulled from public reporting (sources at the bottom).

1. Compress months of engineering into a day

The headline benchmark isn't a benchmark. Anthropic reports Stripe used Fable 5 to migrate a 50-million-line Ruby codebase in a single day — work estimated at two months for a team doing it by hand. That's the kind of throughput that changes what you even attempt.

2. Orchestrate hundreds of subagents from a script it writes itself

Claude Code's dynamic workflows let it write an orchestration script for your task and run it across dozens to hundreds of background subagents. Codebase- wide audits, giant migrations, cross-checked research — tasks too big for one conversation to coordinate now have a coordinator.

3. Rebuild a working app from a screenshot

Fable 5 is state-of-the-art on vision. Anthropic shows it reconstructing a web app's source code from screenshots alone — no markup, no repo, just pixels. "Make it look like this" stops being a metaphor.

4. Read the chart, not just the caption

Same vision leap, quieter use case: it can extract precise numbers from scientific figures and plots. Pulling data out of a PNG of a graph — the bane of every literature review — is now something you can just ask for.

5. Stay coherent across millions of tokens

Fable 5 "stays focused across millions of tokens in long-running tasks," and Anthropic reports its persistent, file-based memory delivers roughly 3× the gains it gave Opus 4.8. Long, multi-day agent runs stop drifting halfway through.

6. Find and fix its own security bugs mid-session

Claude Code's security-guidance plugin reviews Claude's changes for vulnerabilities as it works — a fast pattern check on each edit, a model review at the end of each turn, and a deeper agentic review on commit/push — and fixes what it finds in the same session. The model writing the code is now also the reviewer catching its mistakes.

7. Pursue a goal across many turns until it's actually done

With /goal and auto mode, Claude Code keeps working across turns until a completion condition holds, with a classifier handling routine permissions in the background. Customers describe Fable 5 as delivering "more capable engineering in fewer turns" — fewer babysitting loops, more finished work.

8. Decline the dangerous stuff — without getting dumber

Fable 5 ships with three classifier-based safeguards: cybersecurity, biology/chemistry, and model-distillation requests fall back to Opus 4.8. The notable part: Anthropic says more than 95% of sessions involve no fallback at all, so you get Mythos-class capability on everyday work and a guardrail only where it matters.

9. Propose genuinely novel scientific hypotheses

This one is the Mythos 5 (lifted-safeguard) sibling, but it's striking: in blinded comparisons, life-science experts preferred its hypotheses ~80% of the time over Opus-class models, and protein-design teams reported ~10× faster work on parts of drug design. Frontier research assistance, not just autocomplete.

10. Actually ship it — everywhere, at a known price

Capability you can't deploy is a demo. Fable 5 is available today on the Claude API, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry, at $10 / $50 per million input/output tokens (double Opus 4.8). The frontier is, as of yesterday, a line item you can put in production.

A note on the hype

Double the price, hard safety fallbacks, and a staged subscription rollout (June 9–23) mean this isn't a free upgrade for everyone — and independent reviewers are already poking at where it underwhelms. Read widely before you re-architect around it. Still: the gap between "yesterday" and "today" here is real, and unusually large.

Sources

Articles

Videos