The financial desks we read, gathered in one place. Stories we are licensed to carry are reproduced in full with the writer credited; everything else is a headline and a short quotation that links straight to the publisher.
MarkTechPost AI
StepFun has released Step 5 Preview as open weights. FP8 and BF16 checkpoints are live on Hugging Face and ModelScope. Step 5 Preview is a sparse Mixture-of-Experts model from StepFun built for long-horizon agentic…
MarkTechPost AI
Perplexity has open-sourced SPACE, the sandbox platform behind every task in Perplexity Computer. The Perplexity SPACE sandbox gives each agent task its own Firecracker microVM. Credentials and outbound traffic are…
MarkTechPost AI
Liquid AI released d1-ultra, a text and image decision model that returns calibrated probabilities in 1 forward pass. It scores 65 on the Artificial Analysis Decision Index and costs $0.042 per million input tokens. The…
MarkTechPost AI
Meta released Muse Spark Open, an 816B-parameter open-weight MoE model with 42B active parameters and a 1M-token context. It leads open models on MCP-Atlas and runs on 2 M5 Ultra Mac Studios. The post Meta Releases Muse…
MarkTechPost AI
NVIDIA released Nemotron 3.5 Super VL, an open 120B-parameter multimodal MoE model with 12B active parameters, 1M-token context and NVFP4 weights that run on 1 DGX Spark, plus 2 Bolt edge models. The post NVIDIA…
MarkTechPost AI
Introduction Teams shipping AI agents at scale keep running into the same pattern. Even agents built on top-tier models like Fable and Astra struggle to carry out complex, end-to-end tasks inside a specific product…
MarkTechPost AI Full text
Aikido Security's Altar-1 compresses GLM-5.3 from 1.51 TB to 328 GB, runs on 4x H200 for on-prem pentesting.
MarkTechPost AI Full text
Perplexity Research pairs RFT with hint-guided self-distillation on GLM 5.2, reducing live tool-call failures by 21.2% between checkpoints.
MarkTechPost AI Full text
Fastino's GLiNER2.5-Decide is a 340M open-weight encoder that returns rule-constrained, scored decisions on CPU for routing and guardrails.
MarkTechPost AI Full text
FLUX 3 Action from Black Forest Labs is a 7B open-weights world action model topping RoboLab-120 at 42.92%.
MarkTechPost AI Full text
BottleCap AI's ThinkingCap-Qwen3.8-27B cuts thinking tokens 37.2% across 12 benchmarks with only 0.86pp accuracy loss. Drop-in for vLLM.
MarkTechPost AI Full text
Contrastive-LM's CLM-8B scores agent actions instead of generating text, running up to 9× faster than TypeSafe's Jev zero-shot.
MarkTechPost AI Full text
Learn how to use TypeSafe AI's Jev model in Python for typed decisions, confidence scoring, and async deployment
MarkTechPost AI Full text
Google's Gemini 3.8 Flash TTS and Flash-Lite TTS add prompt-based voice design, 2,000+ voices, and 100+ language support.
MarkTechPost AI Full text
NVIDIA's Nemotron 3 Diarization is a 100M-parameter open-weight model that tracks 8 overlapping speakers in real-time streaming audio.
MarkTechPost AI Full text
Nokia's open-source AnyJev turns open LLMs into calibrated decision models with no training, lifting automatable traffic 6.8x
End of the wire
Stories marked Full text are reproduced here under the terms their publisher allows, with the writer credited and the original linked. Everything else is a headline and a short quotation that links straight to the publisher. How we quote · Our sources
