Paper Instruments Ultramarine 1 Flash: open model for knowledge work
At a glance
Paper Instruments launched Ultramarine 1 Flash for professional knowledge work (finance, law, consulting, ops)—API at $0.15/$0.50 per 1M in/out, claims Fable-class JobBench 62.9 and ~60x cheaper than Fable per task; delivers Excel/PPT/Word and ERP/MES workflows; weights coming soon on Hugging Face, training code MIT. This workflow is listed under open-model, knowledge-work, document-agents, pricing, post-training.
Agents / tools
Ultramarine 1 Flash, Paper Instruments, document agents
Tencent Hunyuan open-weights Hy-Memory-30B-A3B (~3B active MoE) to replace the general LLM on the write path of Mem0/MemOS/Hy-Memory—trained with on-policy distillation + attributed memory RL; beats Claude Opus 5 / DeepSeek V4.1 Flash on a fixed cross-harness memory-writer matrix; Apache-friendly HF weights + GitHub framework.
Parallax is a Windows/Linux desktop workspace that runs Claude Code, Codex, Antigravity or any shell in named panes with a built-in editor, local Git, session directory, voice chat, and saved layouts—agents stay independent (no auto-orchestration); local-only data; subscription required ($11.99/mo or $99.99/yr USD) plus your provider plans.
DeepSeek shipped standalone Search (`POST /v1/tools/search`) and Fetch (`POST /v1/tools/fetch` → Markdown) APIs for agents/RAG: $0.75 per 1,000 successful requests each (empty/failed free); Claude Code built-in Web Search and Codex `web_search` waive Search fees for the first two weeks after launch, then standard pricing—docs live under Search & Fetch on api-docs.deepseek.com.
IBM Granite Switch 4.1 (Oct 15, HF blog): 8B/30B compose chat + Decision aLoRA + Guardian safety aLoRA + Granite-Speech-5.0-Turbo-CTC ASR in ~base VRAM via Activated LoRAs with shared KV cache—swap chat/decision/speech on one endpoint; vLLM plugin and llama.cpp support; community can mix Granite Libraries adapters.