Posts tagged “open-weights”

• Reflection releases Beam’s weights for custom coding assistants

  Beam’s downloadable weights let businesses customize and run the coding model themselves. Its 273 GiB compressed download still needs substantial server capacity.

• Meta releases Muse Spark Open weights for custom AI assistants

  Muse Spark Open gives businesses downloadable AI weights. Its hardware demands, hosted preview and custom licence shape how they can use it.

• Reflection Beam Enters Waitlisted Beta; Open Weights Planned for October

  Reflection’s Beam beta supports text-based coding and agents, but Apache-2.0 weights are pending. What builders can evaluate now—and what must wait.

• Kolibri-1 Self-Hosting: GPU Requirements and German-English Evaluation

  Assess Kolibri-1 for private German-English assistants: FP8 versus BF16 GPU needs, context limits, vLLM compatibility and pre-deployment checks.

• Holo4: The Case for Routing Computer Use by Workflow

  H Company’s Holo4 combines GUI, code and API actions. Its split licenses, public traces and long-workflow results change how builders should deploy it.

• DeepSeek V4 Flash Vision Is Open. The API Is Still Harder to Beat

  DeepSeek released MIT-licensed V4 Flash Vision weights. Here is the cost, deployment reality, benchmark caveats, and agent-builder playbook.

• Cohere’s North Micro Vision Gives Documents Their Pixels Back

  Cohere’s 2.4B North Micro Vision preserves native page detail for local OCR, but builders must supply the runtime and reasoning layer.

• LFM2.5-VL-3B Gives the Edge Eyes. Keep Its Hands Tied.

  Liquid AI’s LFM2.5-VL-3B brings screen grounding, OCR, and tool calling to local devices—but builders should separate sight from action.

• Meta Muse Glimmer Fits a 30B Agent Stack on 24 GB—With Fine Print

  Meta’s open-weight Muse Glimmer 30B targets 24 GB hardware with vision, long context, tools, and DFlash. Here is what builders should test first.

• JoyAI-Video-Edit Hits 30 FPS. The Stream Still Runs on Five Clocks.

  JD’s open-weight JoyAI-Video-Edit reports 30 FPS on one B200. Here’s what latency, stream state, and single-session serving mean for builders.

• DeepSeek V4-Flash-0731 Makes the Harness Part of the Model

  DeepSeek V4-Flash-0731 adds major agent gains, Codex-ready Responses API support, and open weights. Here is how builders should evaluate it.

• LG’s K-EXAONE 2.0 Makes Open Weights a Cluster Procurement Decision

  LG’s 750B K-EXAONE 2.0 is Apache 2.0, agent-ready, and built for eight H200s. Here is what its open-weight economics mean for builders.

← Back to blog