Latest Updates & Release Intelligence
Every entry distinguishes official vendor documentation updates and company announcements from UltraMythos independent analysis.
Claude API workspace headers & token pricing structure verified in platform release notes
Anthropic's platform documentation details the anthropic-workspace-id response header and formalizes per-million-token API rates alongside deprecation schedules for legacy Claude 3.5 and 3.7 snapshots.
Gemini 2.5 Pro 1M-token context window rate limits & batch pricing updated
Google's developer documentation outlines 1-million-token unified context window handling across audio, video, image, and code inputs, with tiered pricing above and below 200K tokens.
Perplexity updates Agent API request fees and Pro credit allocation rules
Independent review of Perplexity's documentation shows how Pro ($20/mo) and Enterprise ($40/seat/mo) plans allocate deep research queries versus pay-per-request API citation calls.
Raycast transitions AI plans from strict rate limits to usage-based AI credit pools
Raycast's pricing model separates standard fast models from frontier reasoning models across Pro, Pro+, and Max tiers while keeping core launcher utilities free.
Unified memory bandwidth scaling from 546 GB/s (M4 Max) to 614 GB/s (M5 Max)
Comparing official Apple technical specifications shows how 40-core GPU configurations increase memory bus throughput by 12.4%, directly accelerating local 70B LLM token decode rates.
OpenAI publishes retirement schedule for legacy o3-mini and GPT-4o snapshots
Production API teams must audit pinned model strings before scheduled snapshot retirements take effect, particularly where structured output schemas depend on specific checkpoint behavior.