GMI Cloud Launches EAGLE-4 Speculative Decoding on B300 Fleet
GMI Cloud is a day-one launch partner for EAGLE-4, reporting 2.7x faster decoding than plain decoding on Qwen3.8-27B across its B300 fleet.
Global AI Digest
-
-
MiniMax M3.1 Flash: Open Weights Coming, 1M Context, 1/3 the Cost of M3
MiniMax confirmed that M3.1 Flash, the anonymous 'Space Bunny' model that topped OpenCode for 12 days, will get open weights soon. It offers 1M context with tex…
-
Muse Wonder Model Series Launches on Cline and Other Dev Platforms
Meta's Muse Wonder model series is now available on Cline, Cloudflare, Cognition, Cursor, OpenRouter, Oracle, and Vercel, with Muse Wonder Full exclusive to Mus…
-
Ornith-1.5-122B Now Free in Command Code
Command Code has made the 122B-parameter Ornith-1.5-122B model available for free, with the model's release claiming frontier-level capability alongside strong…
-
Ornith-1.5-122B Now Free in Command Code
Command Code has made Ornith-1.5-122B available for free to all plans and subscribers, featuring 125B parameters, 262K context, and zero data retention by defau…
-
OpenRouter: Agentic Usage Up 34x Since February 2026, Now 8x Human Usage
OpenRouter reports that agentic usage on its platform has grown 34x since the start of February 2026 and is now roughly 8x higher than human usage on a weekly b…
-
Cline Desktop Adds Plugin System and Goal Plugin for Task Verification
Cline Desktop introduces a TypeScript-based plugin system with slash commands, custom tools, lifecycle hooks, and dynamic rules, plus a Goal plugin that keeps t…
-
Qwen 3.8 Max Scores 96.1% on Tīrtha Tool-Calling Bench at a Quarter of Opus 5.5's Cost
TirthaAI reviewed Qwen 3.8 Max on its tool-calling benchmark using GMI Cloud credits: 96.1% on 800 tasks with reasoning off, 100/100 trick requests passed, and…
-
LiteLLM Adds Unified Access to Seedance 2.5, FLUX 3 Max, MiniMax H3 via fal and Runway
LiteLLM now lets developers call video generation models Seedance 2.5, FLUX 3 Max, and MiniMax H3/H3 Max through a single interface, including models hosted on…
-
MiniMax M3.1 Flash Tops Parallel Web System Search Efficiency Leaderboard
MiniMax M3.1 Flash ranks #1 for efficiency on Parallel Web System's search capabilities leaderboard, matching DeepSeek V4.1 Flash's web research performance at…
