Four reproducible parser failures in vLLM 0.26.0, 0.27.1, and 0.28.0 can drop or garble tool calls and reasoning content while still returning HTTP 200, according to the Ingot team's tests on CPU with…
Vercel's AI Gateway is offering MiniMax M3 and M2.7 models free through GMI Cloud until Sunday, September 6, using the -free model IDs (minimax/minimax-m3-free or minimax/minimax-m2.7-free). After the…
MiniMax, founded by Yan Junjie, is offering 14 days of unlimited access to four models—M3, M2.7, Speech 2.8, and Music 3.0—through GMI Cloud, with the promotion running through September 6. The deal g…
MiniMax hinted at its M3 model running on SambaNova hardware in an undated post, but the linked SambaNova demonstration identifies the model as M2.7 and provides no M3 benchmark results. MiniMax relea…
OMLX, a new local AI server for macOS, persists the KV cache to disk in safetensors format, allowing AI agents to resume without recalculating tens of thousands of context tokens after a server restar…
MiniMax, led by founder, chairman and CEO Yan Junjie, posted an undated workout joke on X pairing its open-weight model pitch with Vercel's v0 AI application builder, promising "Mini chest maxxing wit…
NagaAI launched an AI aggregator platform offering a unified OpenAI-compatible API for chat, image, embeddings, and more, with pricing up to 50% lower than direct provider rates on selected models. Th…
A developer known as Nokka (นก-กา) describes a 'Knowledge Stack' system that combines Obsidian, Hermes Agent, and MiniMax M3 to create a self-improving 'second brain.' The stack addresses common PKM p…
A Vetted Consumer analysis of 27 recent AI models found that 13 ship under Apache 2.0 and 7 under MIT, while the remaining 7 use five custom licenses with varying restrictions, including revenue thres…
In 2026, open-weight models such as Kimi K3, DeepSeek V4, Inkling, and Nemotron 3 advertise million-token context windows, made feasible on consumer hardware by rebuilding attention mechanisms. A 48GB…
DeepSeek released DeepSeek-V4-Flash-0731 on July 31st, an MIT-licensed model with API prices of $0.14 per million cache-miss input tokens and $0.28 per million output tokens, aiming to lower cost barr…
DeepSeek released DeepSeek-V4-Flash-0731, a model that Artificial Analysis ranks ahead of MiniMax M3, a 428B model, and at $0.14 per million input tokens and $0.27 per million output tokens it may be …
Every notable open-weight frontier model released in the past year uses a Mixture-of-Experts (MoE) architecture, according to an analysis by Vetted Consumer. The shift means total parameters range fro…
A developer created a shell wrapper called 'synclaude' that routes Claude Code's interface through Synthetic's API, allowing users to run open models like Kimi K3, GLM 5.2, and MiniMax M3 while keepin…
MiniMax M3, Z.ai's GLM-5.2, and Moonshot AI's Kimi K3, three open-weight models released in mid-2026, compete with closed frontier systems on agentic coding but differ in hardware requirements, latenc…
MoonMath AI launched Zro, a private inference endpoint for coding agents that serves open-weight models from EU infrastructure with zero request retention and no training on customer data. The service…
Together AI launched Provisioned Throughput, a new inference service offering reserved capacity for open-weight models with token-based pricing and a 99% uptime SLA, available for MiniMax M3 and GLM-5…
DeepReinforce released Ornith-1.0, an open-source family of coding models that learn to build their own task scaffolds, moving intelligence from the surrounding harness into the model itself. The mode…
Cline launched ClinePass, a subscription plan offering discounted access to open-weight models including GLM-5.2, Kimi K2.7 Code, and DeepSeek V4 Pro. The plan costs $4.99 for the first month and $9.9…
DeepReinforce open-sourced Ornith-1.0, a family of self-improving coding models ranging from 9B to 397B parameters, which learn to generate their own task-specific scaffolds during reinforcement learn…