Anthropic releases Claude Fable 5.2 and Mythos 5.2 at a lower price than the 5.1 models
Anthropic has released Claude Fable 5.2 and Claude Mythos 5.2, the same underlying model, with Mythos limited to organisations in its cyber and life sciences verification programmes. Both cost $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for the 5.1 models, with a 1M token context window. The release notes warn that requests which edit an earlier prefix and send back 5.2 thinking blocks now return a 400 error.
Meta releases Muse Ember, a small model for agent subtasks priced at $0.10 per million input tokens
Meta Superintelligence Labs has released Muse Ember, its smallest model, built for the repetitive steps of agent runs such as tool calls, coding subtasks, routing and document parsing. It costs $0.10 per million input tokens and $0.50 per million output tokens, with a 1M token context window, on the Meta Model API. Meta says it beats Claude Haiku 5.5 and GPT-6 Luna on intelligence per dollar, citing Artificial Analysis scores alongside its own evaluations.
Google adds TPU 8tx and TPU 8i Max systems and open-sources TPU Sync for KV cache transfer
Google has added two eighth-generation TPU systems. TPU 8tx is for confidential inference that keeps data and model weights protected while in use, and TPU 8i Max packages two TPU 8i chips together for double the compute and HBM bandwidth. It also open-sourced TPU Sync, a library that tiers KV cache into host memory and storage, which Google says cuts KV transfer latency by more than 90%.
AWS launches Nitro Sandboxes to run pods, tasks and AI agents in hardware-isolated VMs
AWS has made Nitro Sandboxes available, lightweight VMs managed by the Nitro Hypervisor that give each Amazon EKS pod, ECS task or small VM its own hardware isolation boundary on a shared EC2 instance. AWS pitches them for running untrusted or LLM-generated code and agent workloads, with cryptographic attestation so secrets are released only to sandboxes that prove their integrity. They cost nothing extra on most Nitro-based instances in all commercial regions.
AWS →
Infrastructure · Agents · Safety
AWS
Vercel's AI Gateway index finds decision models took about a quarter of October requests
Vercel's monthly AI Gateway Production Index says decision models, which classify or score rather than generate text, received about one in four gateway requests in October but only 0.53% of spend, and nearly a third of paid teams used one. Anthropic was first choice for nearly twice as many new teams as OpenAI, though its share of spend has roughly halved since December to about 32 cents per dollar. The figures cover Vercel's customers only.
Vercel →
Models · Funding
Vercel, Anthropic, OpenAI