Perplexity said on March 12, 2026 that Computer is now available to Pro subscribers, widening access beyond its highest tier. The company is pitching 20+ advanced models, prebuilt and custom skills, and hundreds of connectors, while reserving monthly credits and higher spend limits for Max users.
On March 13, 2026, GitHub showed a Copilot SDK and Remotion demo that turns a WhatsApp message into a promo video in about five minutes. GitHub’s official SDK announcement describes the stack as a programmable layer that can plan, invoke tools, edit files, and run commands inside other applications.
GitHub used X on March 15, 2026 to spotlight the Copilot CLI `/fleet` command for routine maintenance work. GitHub’s official Copilot CLI materials now describe `/fleet` as a parallel sub-agent workflow that converges multiple runs into one decision-ready result.
A LocalLLaMA release post presents OmniCoder-9B as a Qwen3.5-9B-based coding agent fine-tuned on 425,000-plus agentic trajectories, with commenters focusing on its read-before-write behavior and usefulness at small model size.
A former Manus backend lead argues that one run(command="...") tool can outperform large typed tool catalogs because CLI patterns fit how LLMs consume text, prompting a debate over flexibility versus sandboxing.
Sebastian Raschka's LLM Architecture Gallery drew attention on HN for turning recent model families into comparable diagrams, making dense, MoE, and hybrid design choices easier to scan in one place.
A high-traffic Ask HN thread shows a polarized view of AI coding tools: developers report clear gains on small scoped tasks, but many say autogenerated specs and cleanup work are eroding team velocity.
OpenAI said on February 20, 2026 that its theorem-proving model produced proof attempts for all 10 research-level First Proof problems. After expert feedback, the company believes at least five attempts are likely correct, while some remain under review and the attempt for problem 2 now appears incorrect.
GitHub moved Copilot’s coding agent for Jira into public preview on March 5, 2026. Teams can assign Jira Cloud issues to the agent, let it implement changes in a connected repository, open a draft pull request, and post progress back into Jira.
Google put Gemini Embedding 2 into public preview on March 10, 2026. The company says the model handles text, images, and mixed multimodal documents in one embedding space while improving benchmark scores to 68.32 for text and 53.3 for image tasks without changing price or vector dimensions.
Google DeepMind updated Gemini 3.1 Flash-Lite on March 3, 2026 as a low-cost model for high-volume, low-latency work. Google says it supports 128k input, 8k output, multimodal input, native audio generation, and pricing from $0.10 per 1M input tokens.
A high-signal LocalLLaMA thread on March 15, 2026 focused on a license swap for NVIDIA’s Nemotron model family. Comparing the current NVIDIA Nemotron Model License with the older Open Model License shows why the community reacted: the old guardrail-termination clause and Trustworthy AI cross-reference are no longer present, while the newer text leans on a simpler NOTICE-style attribution structure.