Know
What changedClaude Fable 5.2 is generally available
Anthropic says Fable 5.2 and Mythos 5.2 share an underlying model; Fable, with cybersecurity and biology safeguards, is generally available across platforms, while Mythos is restricted to approved program participants. Anthropic says generation is up to 30% faster than 5.1. Standard pricing is $8 per million input tokens and $40 per million output; optional fast mode costs $16/$80.
Compare Fable’s standard pricing with your current model; check platform-specific terms before adopting fast mode.
Meta launches low-cost Muse Ember
Meta released Muse Ember for tool calls, coding subtasks, routing, extraction and long-document parsing, with access through Meta Model API and Muse Code. It accepts text, images, audio, video and PDFs, with a 1-million-token context and up to 512,000 output tokens. Pricing is $0.10 per million input tokens and $0.50 per million output tokens. Meta’s benchmark and cost-per-task comparisons are its own claims.
Test Ember on high-volume routing or extraction jobs, and measure quality against your current model.
Fireworks reports broader incident impact
Fireworks says its October incident affected account and billing information and, for a limited set of accounts, secrets, datasets and incomplete model weights. The company attributes the access to a compromised third-party cloud provider, followed by access to Fireworks development machines. It reports no impact to Inference, Training SDK or Nexus; that is the company’s assessment.
Check Fireworks’ incident notice and your account exposure; rotate any credentials or secrets the company identifies as affected.
ChatGPT’s dot can use multiple computers
OpenAI says a single dot, its always-on ChatGPT agent, can now connect to multiple local computers and select a machine for each task. Its example is building and testing an app on a dedicated Mac while you continue working on a laptop. The captured summary does not specify eligibility, setup steps or pricing.
If you use dot, check whether multi-computer access is enabled and try delegating a test run to a separate machine.
Gemini may expose connected Gmail and Drive content to review
A summary of Google’s help documentation says personal-account users with Workspace connected and Keep Activity on may have some Gmail and Drive content reviewed by humans. That can include summaries, excerpts, inferences and sometimes full messages or files; reviewed data may be retained for up to three years and cannot then be deleted. The page itself was not independently captured.
If this applies to your account, review the Gemini settings and disconnect Workspace or turn off Keep Activity to stop future review, per the summary.
Google expands its eighth-generation TPU lineup
Google’s TPU 8tx uses TPU 8t silicon and fifth-generation AMD EPYC hosts, and is designed for confidential inference and small-scale training. TPU 8i Max pairs two TPU 8i chips, doubling compute, SRAM and HBM bandwidth versus TPU 8i, with massive context windows and rack density as targets. The captured summary gives no pricing or availability details.
Track cloud availability and benchmarks before planning deployments around either new TPU system.
ChatGPT Enterprise adds AWS Marketplace purchasing
OpenAI made ChatGPT Enterprise available through AWS Marketplace, so organizations can buy and manage it through existing AWS billing and apply eligible spend toward AWS commitments. The captured summary says AWS customers receive the same models and features as direct buyers. ChatGPT for Healthcare and Financial Services are also available to eligible customers.
Ask your procurement team whether Marketplace billing can use existing AWS commitments.
Anthropic reports working electronics prototypes
Anthropic says Claude Fable 5.2 worked with circuit-board maker Diode Computers to design a speakerphone, voice-assistant hub and handheld virtual pet. It laid out boards, modeled designs in CAD and accounted for manufacturing constraints; the company says the prototypes worked without circuitry fixes. This is a reported research result, not a general-purpose design tool release.
Treat this as a signal of expanding model-assisted hardware design, not as a validated workflow for production.