Faster inference variant of Kimi's coding specialist
Added Jun 12, 2026
Kimi K2.7 Code Highspeed is the high-speed serving variant of Moonshot AI's K2.7-Code model, released June 12, 2026. It delivers roughly 180 tokens per second for coding tasks while preserving the same 256K context window, text/image/video input, and thinking-mode capabilities.
Why: Kimi K2.7 Code Highspeed is the latency-optimized version of an already strong coding model, making it a good pick for interactive coding agents and live pair-programming workflows.
Limited-availability Mythos-class model without Fable 5 safety classifiers
Added Jun 9, 2026
Anthropic's Mythos-class model announced on June 9, 2026, shares the same capabilities as Claude Fable 5 without the safety classifiers. It is offered only in limited availability to approved customers through Anthropic's Project Glasswing program.
Why: Mythos 5 is a notable limited-availability variant of the Mythos-class tier, distinct from the generally available Fable 5.
Google DeepMind open-sourced DiffusionGemma, a 26B MoE text-diffusion model that generates text up to 4x faster than comparable autoregressive models for on-device AI workloads.
Cohere open-sourced North Mini Code, a 30-billion-parameter MoE coding agent that runs on a single H100. Independent tests show it can triple output tokens versus comparable models, a verbosity tradeoff teams must budget for in high-volume agent pipelines.
Mastercard unveiled Agent Pay for Machines, a protocol letting AI agents send micropayments and settle with each other, joining a wave of big-company moves to build payment rails for agentic commerce.
Google released Gemini 3.5 Live Translate for near-real-time speech-to-speech translation across 70+ languages, rolling out through Google Meet, Translate, and the Gemini Live API with SynthID audio watermarks.