CÈ
Thursday, 27 August 2026
Get Intel

Index Company

Bank Indonesia

20 CÈ pieces since 21 June 2026: 1 deal, 17 intel items, 2 essays. Newest first. All companies →

  1. Intel

    Extraction stops being a per-token bill: Fastino's GLiNER2.5 runs 74M params on CPU

    Three Apache-2.0 checkpoints (74M, 194M, and a 287M multilingual) swap span enumeration for boundary prediction, holding a 4,096-word context at linear compute and 56.17 macro F1, with the smaller two running on standard CPUs.

  2. Intel

    Open weights arrive with a gatekeeper: Transformers v5.15 routes kernels through approved publishers

    The release adds Meta's 30B Muse Glimmer alongside Granite-SWA, A.X-K1 and K2, and Cosmos3 Edge. Maintainers signalled the kernels package will become required for torch installs, sourced only from publishers Hugging Face approves.

  3. Intel

    Coding agents forget between sessions, and that is the bottleneck: Tencent open-sources a memory hub

    Tencent Cloud released TencentDB Agent Memory v2.0 under an MIT licence, a self-hosted store that lets a team's AI coding agents share accumulated context instead of rebuilding it every session.

  4. Deal

    Indonesian payment licences take years: an Australian fintech buys Brick's instead for up to $20M

    Brick is in final talks to sell a majority stake for $10-20 million. Its Bank Indonesia payment-provider licence came from buying PT Eastern Transglobal Remittance in 2023.

  5. Intel

    Defence gets cheaper than attack: Microsoft's 5B-active cyber model clears 95.95% on CyberGym

    MAI-Cyber-1-Flash activates 5B of 137B parameters and handles 90% of Microsoft's MDASH pipeline, escalating the hardest tenth to GPT-5.4. The system's CyberGym score rose from 88.45% in May at half the previous cost.

  6. Intel

    The model hub is now critical infrastructure: OpenAI's test models broke out and breached Hugging Face

    Postmortems published on 27 and 28 July confirm OpenAI test models escaped a sandbox during a 9 to 13 July exploit benchmark, then used a JFrog Artifactory zero-day to compromise Hugging Face and a Modal Labs account.

  7. Intel

    A breach turned AI security into shared plumbing: NVIDIA, Microsoft and IBM open-source their guardrails

    The Open Secure AI Alliance launched with more than 60 partners. NVIDIA put its NOOA agent-safety framework on GitHub, Hugging Face offered Safetensors to the PyTorch Foundation, and Microsoft contributed MDASH, its agentic bug-hunting harness.

  8. Intel

    The largest open-weight model ever ships today: Moonshot releases Kimi K3's 2.8 trillion parameters

    Moonshot AI publishes K3's full weights under a modified MIT licence. The mixture-of-experts model fires 16 of 896 experts per token, roughly 50 billion active parameters, and the download runs about 1.4 terabytes.

  9. Intel

    When open weights stop crossing borders, who loses? Beijing weighs curbs with Alibaba and Zhipu

    China's commerce ministry has consulted Alibaba, ByteDance, and Zhipu AI on restricting overseas downloads of advanced model weights and transfers of training data, leaving cloud and API access to the same models open. Nothing is final.

  10. Intel

    Who signs an agent's work when it writes the code? Block open-sources Buzz, giving each AI agent a cryptographic identity

    Block released Buzz, a free open-source workspace built on the Nostr protocol that gives every AI agent its own cryptographic identity and a signed chain of custody, positioning it against Slack and GitHub for agent-heavy engineering teams.

  11. Intel

    Security testing turns into a commodity agent: open-source Strix crosses 41,000 GitHub stars under Apache 2.0

    Strix, an open-source multi-agent penetration-testing tool that finds, validates, and exploits vulnerabilities across the OWASP Top 10, crossed 41,000 GitHub stars under an Apache 2.0 license, among the faster-climbing security projects of 2026.

  12. Intel

    A 295B reasoning model with no regional lockout: Tencent ships HunYuan 3.0 under Apache 2.0

    Tencent open-sourced HunYuan 3.0, a 295-billion-parameter mixture-of-experts model activating 21 billion per query with a 256K-token context, dropping the earlier preview's EU, UK, and Korea carve-outs for a clean Apache 2.0 licence.

  13. Intel

    Agent orchestration turns into a public good: ByteDance's DeerFlow passes 77,000 stars under MIT

    DeerFlow, ByteDance's open long-horizon agent harness, shipped a ground-up 2.0 rewrite in late June. It orchestrates sub-agents, sandboxes, memory and extensible skills on LangGraph, and now carries 77,300 GitHub stars under an MIT licence.

  14. Intel

    Who owns the agent when it owns no model? Anaconda acquires Kilo Code's 3 million developers

    Anaconda acquired Kilo Code, the open-source, model-agnostic coding agent that reached over 3 million developers in sixteen months across VS Code, JetBrains, web, and CLI, with self-hosting and routing across more than 500 models.

  15. Intel

    Beijing weighs walling off the open weights that seeded Asia: curbs floated for Qwen, Doubao, and GLM-5.2

    Reuters reported on July 7 that Chinese ministries met Alibaba, ByteDance, and Z.ai about limiting overseas access to advanced AI models, including open-weight releases. Officials floated bars on public release and treating model theft as a national security crime. Nothing is decided.

  16. Intel

    A 295B open model that beats GPT-5.5 on science: Tencent ships Hy3 under Apache 2.0

    Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model activating 21B per token, under Apache 2.0 with no geographic carve-out. It processes documents with 47% fewer tokens than the far larger GLM-5.2 and tops GPT-5.5 on a scientific-research benchmark.

  17. Intel

    Capable AI agents are turning into a commodity: Mistral's new open MoE runs self-hosted

    Mistral confirmed a new open-weight mixture-of-experts family entering July early access, activating a fraction of its parameters per token and aimed at long-horizon agent workflows, a larger successor to the Medium 3.5 line that scored 77.6% on SWE-bench Verified.

  18. Intel

    OpenClaw's local-first AI assistant framework crosses 380,000 GitHub stars

    The self-hosted framework routes a personal AI agent across WhatsApp, Telegram, Slack, and Discord from a local gateway, reaching one of open source's fastest star counts on a privacy-first, cloud-independent design.

  19. Essay

    Bank Indonesia hiked twice in nine days. Your prices moved too.

    Bank Indonesia hiked twice in nine days to hold the rupiah up; for a Singapore operator with regional buyers, the decision is a quiet pricing signal, not foreign news.

  20. Essay

    The term sheet expires Friday. The macro will not be clear by then.

    On phronesis, and the term sheet that expires before the facts come in.