AI Product Updates Daily — June 10, 2026
Anthropic opens its Mythos-class AI to the public with Claude Fable 5: 80.3% on SWE-Bench Pro, $10/$50 per million tokens, available on Claude API, Bedrock, Vertex, GitHub Copilot, and Snowflake. Apple's WWDC developer session makes Private Cloud Compute free for smaller developers and adds a multi-model Swift API that routes to Claude or Gemini.
The big story of June 9 was Anthropic finally bringing its Mythos-class capability to the public — four months after the technology first appeared in restricted cybersecurity circles. Alongside it, Apple used its developer conference to turn the Foundation Models Framework into a proper AI platform, complete with free cloud compute for smaller developers and a Swift API that can now route calls to Claude, Gemini, or any third-party model. Two high-conviction moves in one day, aimed at very different audiences.
Anthropic: Claude Fable 5 and Mythos 5
Claude Fable 5 is the first Mythos-class model Anthropic has made available to the general public, released June 9.1 Until now, Mythos-tier capability existed only inside Project Glasswing — a restricted program for vetted cybersecurity operators, critical-infrastructure organizations, and select biology researchers. Fable 5 is the same underlying model as Claude Mythos 5, with an additional layer of hard safety guards for public use.
What Fable 5 can do
Anthropic positions Fable 5 for long-horizon, autonomous tasks: large codebase migrations, multi-stage agent workflows, deep research, and complex document analysis. The company's own framing: the longer and harder the task, the wider Fable 5's lead over any earlier model.2
Selected benchmark figures published at launch:
| Benchmark | Fable 5 / Mythos 5 | Claude Opus 4.8 | GPT-5.5 |
|---|---|---|---|
| SWE-Bench Pro | 80.3% | ~69% | 58.6% |
| Legal reasoning | 13.3% | lower | 2.1% |
| Spatial reasoning | 38.6% | 14.5% | — |
(Scores are the higher of the two model configs. Starred rows — cybersecurity, biology, chemistry — reflect Mythos 5 figures; on those tasks Fable 5's live performance, with safeguards active, falls closer to Opus 4.8.)
Stripe ran a 50-million-line Ruby migration using Fable 5 and completed it in a day — work the company estimated at more than two months for a full engineering team.3 On FrontierCode (Diamond), a benchmark for production-codebase difficulty tasks, Fable 5 leads even at medium reasoning effort, which keeps token costs in check. Spatial reasoning nearly tripled versus Opus 4.8 (38.6% vs. 14.5%).
Full benchmark comparison from Anthropic's launch materials 2
Andrej Karpathy's take on launch day:
"The benchmarks are great and it's SOTA on everything by a margin but I'll add that qualitatively also, this is a major-version-bump-deserving step change forward... peaking especially for long problem-solving sessions on very difficult problems. You can give it a lot more ambitious tasks than what you're used to, the model 'gets it' and it will just go..."
He also flagged that the launch-day safeguards are "configured to be a little too trigger happy" — something Anthropic acknowledged and said it will tune.
Pricing and availability
Fable 5 is priced at $10/M input tokens and $50/M output tokens, exactly double Opus 4.8 — but less than half what Claude Mythos Preview cost under the earlier restricted program.4
| Plan type | Access |
|---|---|
| API / consumption-based Enterprise | Available today, no cut-off date |
| Pro, Max, Team, seat-based Enterprise | Included free through June 22 |
| Post-June 23 subscriptions | Requires usage credits; Anthropic plans to restore it to standard subscription tier as capacity allows |
Prompt caching gives a 90% discount on cached inputs; US-only inference is available at 1.1× pricing for data-residency needs. The API model string is
claude-fable-5.Fable 5 is live on Claude API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, GitHub Copilot (Pro+, Max, Business, Enterprise), and Snowflake Cortex AI.56
The safeguard fallback
Because Mythos-class capability carries elevated misuse risk — internal Anthropic evaluations showed the model could operate at a high level in vulnerability discovery and exploitation — Fable 5 ships with hard classifiers covering cybersecurity, biology/chemistry, and model distillation. When a request trips one, the model silently hands off to Opus 4.8 and notifies the user. Early data shows this happens in fewer than 5% of sessions.4
One consequential policy shift: all Mythos-class traffic, even for Enterprise customers who previously had zero-retention agreements, now requires 30-day data retention. Anthropic says the data is used only to defend against novel jailbreaks and identify false positives, not for training, and is deleted after 30 days. The policy may establish a precedent for how the industry handles access to high-capability models.
Claude Mythos 5 is available only to users already approved under the previous Mythos Preview program (Project Glasswing cybersecurity partners and select biology researchers). Anthropic says it plans to expand that via a trusted-access program, including future US government partnerships.
Apple: Foundation Models and Core AI become a developer platform
Apple's WWDC 2026 Platforms State of the Union on June 9 went further on the developer-tools side than the keynote had suggested.7
Foundation Models Framework expansion
The headline change for developers: free Private Cloud Compute access for apps with fewer than two million first-time App Store downloads. The previous cost barrier — paying for inference on Apple's secure cloud infrastructure — is gone for the majority of indie developers.
Beyond pricing, the framework gained:
- Image input support — on-device and cloud requests can now include visual content
- Server-side model integration — a new
LanguageModelprotocol routes API calls to Apple's own models or, via the same Swift API, to Claude, Gemini, or any other provider the developer chooses, with no code changes required to swap - Dynamic Profiles for multi-agent workflows — structured support for building chains of agents within the framework
- Open source release, scheduled for later this summer
Core AI framework
Apple introduced Core AI as a brand-new framework for running fully custom on-device models. It ships with ahead-of-time compilation, dedicated Instruments support, and Python tooling to convert PyTorch models to Apple silicon. This is the layer powering Siri internally and is now exposed to third-party developers.7
Xcode 27
Xcode 27 is Apple silicon-only, 30% smaller than its predecessor, and expands agentic coding significantly: agents can now interact with the iOS Simulator, localize apps, run test suites, and pull crash reports from Organizer to fix them automatically. Xcode Cloud builds are up to twice as fast.
A new View Annotations API lets Siri act on whatever is currently on screen in a conversational loop, and new App Intents entity/intent schemas make app content indexable and addressable through natural language.
Siri AI — what the developer session confirmed
For context from the June 8 keynote: Apple's rebuilt Siri AI runs on Google Gemini models via Private Cloud Compute, has a standalone app across all Apple platforms, and supports full agentic task execution.8 Developer beta is live as of June 9. Public beta follows in July; the full release is scheduled for fall 2026.
Two caveats noted by multiple outlets: the new Siri AI will not launch in the EU due to DMA compliance requirements Apple says it cannot meet without opening equivalent data access to competitors. And the highest-end Siri AI features require the latest Apple silicon — the iPhone 16 family, despite being marketed as "built for Apple Intelligence," does not meet the threshold.
Intel deprecation, Liquid Glass, and MLX
macOS Tahoe was the last Intel-compatible release. Developers can now ship Apple silicon-only binaries on the Mac App Store. Apps recompiled with Xcode 27 will automatically adopt the Liquid Glass design language, with no opt-out. MLX, Apple's open-source ML research framework, now supports Metal 4 and can scale model training across multiple Macs over RDMA/Thunderbolt.
Other updates
Lovable reached $500M annualized revenue as of June 9, with one million new projects started per week — a data point on the pace at which AI-assisted software building is converting into commercial volume.9
No new model releases from OpenAI, Google, xAI, or Microsoft were dated June 9–10. The LLM Stats changelog shows Claude Fable 5 as the sole model release on June 9.10
References
- 1Anthropic — Introducing Claude Fable 5 and Claude Mythos 5
platform.claude.com
- 2
- 3Claude Fable 5 API, Benchmarks, Pricing — TrueFoundry
truefoundry.com
- 4
- 5
- 6
- 7
- 8
- 9
- 10AI Updates Today — LLM Stats
llm-stats.com
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.