ScaleDown's Constrained-Decoding Models Launch on OpenRouter
ScaleDown's family of constrained-decoding models is now live on OpenRouter, with task-specific models for summarization, extraction, and classification, plus l…
Global AI Digest
-
-
Requesty Partners with Melious to Route GLM-5.3 and DeepSeek V4 Models via European Infrastructure
Requesty announced a partnership with Melious to offer European-routed inference for GLM-5.3, GLM-5.3 Flash, DeepSeek V4.1 Flash, and DeepSeek V4 Flash through…
-
Meta's Muse Wonder Now Available on Requesty Gateway
Meta's Muse Wonder model is live on Requesty, designed for long-running agentic tasks with cross-session memory and priced at $4/$20 per 1M tokens with 1M conte…
-
Tencent AI Introduces LightVela: A 24/7 Cloud AI Assistant for Telegram, WhatsApp, and Discord
Tencent AI announced LightVela, a cloud-based AI assistant that runs 24/7 and integrates with Telegram, WhatsApp, and Discord. Users can create bots in seconds…
-
MiMo-V2.6-Pro Tops Open-Weight Models on FrontierCode 1.1 with 45%
Xiaomi's MiMo-V2.6-Pro leads all open-weight models on Cognition's FrontierCode 1.1 benchmark with a 45% score, a private evaluation testing whether models can…
-
B.AI Reports 1.51T Daily Token Throughput and 70+ Model Matrix
B.AI disclosed infrastructure figures including a 70+ model matrix, a unified Responses API, and 1.51 trillion tokens processed in a single day.
-
Hugging Face Kernels Cookbook Released
A new cookbook covering how to use Hugging Face kernels from the Hub, hardcoding Triton kernels, and benchmarking models with and without kernels.
-
Tencent AI Releases ExplorationBench: A Benchmark Testing Whether AI Can Explore, Not Memorize
Tencent AI introduces ExplorationBench, a benchmark built on two 'alien worlds' where learned knowledge does not apply. Ten frontier models scored below 16% bef…
-
Mistral Large 4 Scores 13.3% on sudo L7 Software Engineering Benchmark
Mistral Large 4 has been added to the sudo L7 leaderboard, scoring 13.3% and outperforming Muse Spark 1.3 and Grok 4.7 on a benchmark designed to test staff-lev…
-
MiMo-V2.5 Series Models Migrate to V2.6 with Scheduled Deprecation
MiMo-V2.5-Pro and MiMo-V2.5 will be auto-routed to MiMo-V2.6-Pro and MiMo-V2.6-Flash starting Oct 14, 18:00 SGT, billed at new rates, and fully deprecated on Oc…
