Sarvam Akshar launched today with zero technical details. After their last model got 334 downloads and was called “embarrassing,” it’s time to ask: Is this sovereign AI or sophisticated vaporware?
Google’s Gemini 3 Deep Think just crushed every AI reasoning benchmark (84.6% ARC-AGI-2) – but at $250/month with 5 prompts/day, is it the AGI preview for the elite or a preview of AI’s accessibility crisis?
AirLLM lets you run Llama 3.1 405B on 8GB VRAM and 70B models on 4GB GPUs through layer-wise inference. Here’s the catch: it’s insanely slow. Is the tradeoff worth it?
OpenAI’s GPT-5.2 derived a new result in theoretical physics, proving that single-minus gluon amplitudes are nonzero. The discovery, verified by…
OpenAI’s GPT-5.3-Codex-Spark is here. Running at 1,000+ TPS on Cerebras WSE-3 hardware, it redefines the speed of agentic coding and “Software 3.0.”
MiniMax M2.5 just dropped with 80.2% on SWE-bench, 100 tokens/second speed, and a price tag that makes intelligence “too cheap…
Canvas-of-Thought replaces linear Chain-of-Thought with mutable DOM-based reasoning. New paper shows it beats CoT, ToT, and PoT on VCode, RBench-V, and MathVista.
University of Michigan’s Prima AI can diagnose brain MRIs in 3 seconds with 97.5% accuracy. This vision-language model trained on 200K+ scans identifies 50+ neurological conditions and could solve radiology’s workforce crisis.
GLM-5’s 744B open-source beast crushes Opus 4.6 on price at $1/1M tokens, but Anthropic’s 1M context window dominates knowledge work. The February 2026 coding showdown explained.
A reference to “Gemini 3.1 Pro Preview” appeared on Artificial Analysis Arena. A Google DeepMind employee hinted “Thursday seems likely.” Here’s what’s happening. [158 chars]
