RSS.bz

AI Daily · Nov 7, 2026: OpenRouter launches Auto Router and Auto Code, with Auto Code at 74% lower cost

OpenRouter launched Auto Router and Auto Code, with Auto Code benchmarking near Opus 5.5 at 74% lower cost and 0.65s p90 latency. Parallel Decisions arrived on OpenRouter with 7 days free, while DeepSeek V4.1 Flash hit Command Code at 1000 TPS.

Pick a date 4/2019: 0 briefings
4/2019
SunMonTueWedThuFriSat
123456
78910111213
14151617181920
21222324252627
282930

Dotted dates have a briefing; click one to open it.

  • OpenRouter launches Auto Router and Auto Code model routers

    OpenRouter introduced next-generation routers: Auto Router for general Q&A and agents, and Auto Code for coding agents, which benchmarks similar to Opus 5.5 at 74% lower cost with 0.65s p90 latency; beta showed 4x fewer invalid tool calls, 50% fewer cut-off agentic responses, and 3x lower latency vs the previous router.

    @OpenRouter 3 duplicate reports merged

  • Parallel Decisions launches on OpenRouter with seven-day free requests

    Parallel Decisions is now available on OpenRouter, offering web-grounded structured answers with probabilities and citations, free for 7 days for OpenRouter users; it claims 13x faster answers at 1/16 the cost.

    @OpenRouter 1 duplicate reports merged

  • DeepSeek V4.1 Flash live on Command Code at 1000 TPS

    DeepSeek V4.1 Flash 1017 Ultrafast is now live in Command Code with a 1M token context window and 1000+ TPS throughput; GLM 5.3 Flash runs at 500 TPS, and the $10/mo GOAT plan includes $20 credits (2x usage).

    @MrAhmadAwais 3 duplicate reports merged

  • Qoder adds stealth model Qoduck with daily free access

    Qoder announced a stealth model called Qoduck, available free every day for all individual users for a limited time. Qoder noted high demand at peak times may cause waits.

    @qoder_ai_ide 2 duplicate reports merged

  • OpenCode Browse launches, letting agents control real browsers

    OpenCode Browse is now available, enabling opencode or any agent to control a real browser including tabs, logins and extensions while keeping everything on device without connecting accounts to a centralized cloud agent.

    @thdxr 2 duplicate reports merged

  • LongCat-2.5 gets one week of free access on Nous Portal

    Meituan LongCat's multimodal model LongCat-2.5 is available for free on the NousResearch Portal for one week, alongside Hermes Agent.

    @Meituan_LongCat

  • Gemini Omni 1.2 Flash video model available on GMI Cloud

    Gemini Omni 1.2 Flash is now available on GMI Cloud, supporting up to 15-second video generation at 30% lower cost, with improved editing and more natural speech. The model ranked #1 in Text-to-Video with 1541 points and #1 in Image-to-Video, and #3 in Video Editing.

    @gmi_cloud 1 duplicate reports merged

  • LiteLLM adds -latest model aliases and Claude limit fallbacks

    LiteLLM now supports -latest aliases that automatically point to newer model versions without code changes, and missing Claude limits fall back to family defaults (#49815); it also lists temperature/top_p only when supported (#50144).

    @LiteLLM 2 duplicate reports merged

  • Hugging Face releases MoE training handbook after $2m spend

    Hugging Face published a handbook on training Mixture-of-Experts LLMs at scale, distilling lessons from a $2 million effort into chapters and figures covering how MoE models are built.

    @huggingface

  • OpenAI and FMF release cyber jailbreak severity framework

    OpenAI partnered with the Forum for Incident Response and Security Teams member firms on a framework to identify when bypassing an AI system's safeguards could enable serious cyber harm and to prioritize stronger protections.

    @OpenAI