AI DAILY / DEV
←
FRIDAY
June 5, 2026
→

    MiniMax Drops M3 — Open-Weight 1M-Context Model Beats GPT-5.5 on SWE-Bench Pro
    • June 1 launch: first open-weight model to combine frontier coding, 1M-token context and native multimodal (text/image/video) input in one architecture.
    • Scores 59.0% on SWE-Bench Pro — ahead of GPT-5.5 (58.6%) and Gemini 3.1 Pro (54.2%), still behind Claude Opus 4.8 at 69.2%.
    • New MiniMax Sparse Attention (MSA) cuts per-token compute to ~1/20 of M2 at 1M context, with 9.7× faster prefill and 15.6× faster decode.
    • API live now at $0.60/$2.40 per 1M input/output tokens — roughly 8–20% of leading proprietary U.S. models; weights and technical report promised within 10 days.
    models minimax.io

    OpenAI Upgrades GPT-Rosalind With GPT-5.5 Agentic Stack — Beats Base Model on Every Life-Sciences Eval
    • June 3 update folds GPT-5.5's agentic coding and tool-use into the life-sciences model, with new plugins for AlphaFold, PubMed and bioinformatics pipelines inside Codex.
    • New LabWorkBench (real, never-trained-on wet-lab protocols): GPT-Rosalind 63.2% vs GPT-5.5 55.8% using 5.3% fewer tokens.
    • On GeneBench long-horizon genomics analysis: 21.6% vs 20.4% accuracy with 31% fewer tokens than GPT-5.5.
    • Research preview now open to eligible orgs worldwide; Novo Nordisk joins Amgen, Moderna, Allen Institute and Thermo Fisher in the program.
    models openai.com

    Autonomous AI Tool Finds 2-Year-Old Redis RCE in Most Cloud Environments
    • CVE-2026-23479 ("RediShell"): use-after-free in Redis blocking-client code lets an authed user run arbitrary OS commands; introduced in 7.2.0 and unpatched until May 5.
    • Found by Theori's Xint Code, an autonomous AI bug-hunter — demo'd as a working RCE at Wiz's ZeroDay.Cloud 2025 in London last December.
    • NVD scores it 8.8 (CVSS 3.1); Wiz's scan puts Redis in a majority of cloud environments, most running without a password, meaning the default user already has every privilege the chain needs.
    • Concrete data point in the week's HN narrative that AI-found zero-days are arriving faster than vendors can patch them.
    research thehackernews.com

    xAI's Grok Imagine Video 1.5 Tops Image-to-Video Arena With 720p Clips and Native Audio
    • June 3 preview API: image-to-video at 720p/24fps, 6–15 second clips, with synchronized dialogue, lip-sync, SFX and ambient audio generated in a single inference pass.
    • +52 Elo on Image-to-Video Arena over Grok Imagine 1.0 — #1 ahead of ByteDance Seedance 2.0, Alibaba HappyHorse 1.0 and Google Veo.
    • API pricing: $0.08/sec at 480p, $0.14/sec at 720p; optimized for clip extension so users can chain shots into multi-shot narratives.
    • Musk posted a Grok Imagine 1.5–generated Iliad movie trailer on X on June 4 captioned 'want full movie?' — story trended across film-tech Twitter.
    models x.ai

    Meta Business Agent Goes Global Across WhatsApp, Instagram and Messenger
    • June 3 Conversations 2026 keynote in London: 2-year limited test ends, agent is now available to every business worldwide; 1M+ businesses already running one on WhatsApp/Messenger.
    • Handles answering questions, booking appointments, qualifying sales leads, recommending products and reroutes to humans when needed; getting started is free.
    • Companion Meta Business Agent Platform connects agents to hundreds of systems (Shopify, Zendesk, etc.) so they can take action on behalf of the business.
    • Pitched as Meta's first real consumer-AI revenue lever after Muse Spark — the agent runs free now, paid tiers arrive 'in the coming months'.
    industry fb.com