---
title: "AlphaSignal Infra News"
url: "https://alphasignal.ai/topic/infra"
description: "Latest AI and machine learning news about Infra on AlphaSignal."
---

# AlphaSignal Infra News

Latest AI and machine learning news about Infra on AlphaSignal.

## Articles

- [Goodfire Cuts AI Jailbreaks From 66 to Zero on Kimi K3](https://alphasignal.ai/news/goodfire-cuts-ai-jailbreaks-from-66-to-zero-on-kimi-k3.md) (2026-10-08)
- [vLLM 0.31 Speeds Up DeepSeek and Slashes Engine Restart Times](https://alphasignal.ai/news/vllm-0-31-speeds-up-deepseek-and-slashes-engine-restart-times.md) (2026-10-08)
- [George Hotz's tinybox Goes Configurable, Starting at $7,000 Without GPUs](https://alphasignal.ai/news/george-hotz-s-tinybox-goes-configurable-starting-at-7-000-without-gpus.md) (2026-10-08)
- [vLLM-Omni Unifies Text, Speech, and Video Serving With 91% Faster Completion](https://alphasignal.ai/news/vllm-omni-unifies-text-speech-and-video-serving-with-91-faster-completion.md) (2026-10-08)
- [GitHub Copilot Routes Tasks Between Local and Cloud Using MAI Code 1.1 Flash](https://alphasignal.ai/news/github-copilot-routes-tasks-between-local-and-cloud-using-mai-code-1-1-flash.md) (2026-10-07)
- [vLLM's Vela 2.0 Collapses Nine AI Safety Checks Into One Open Model](https://alphasignal.ai/news/vllm-s-vela-2-0-collapses-nine-ai-safety-checks-into-one-open-model.md) (2026-10-07)
- [PrismML's Ternary Bonsai 2 Runs 262K Context on a 12 GB GPU](https://alphasignal.ai/news/prismml-s-ternary-bonsai-2-runs-262k-context-on-a-12-gb-gpu.md) (2026-10-07)
- [Mia AI Lab's sparkDash Watches Your Entire NVIDIA DGX Spark Fleet at Once](https://alphasignal.ai/news/mia-ai-lab-s-sparkdash-watches-your-entire-nvidia-dgx-spark-fleet-at-once.md) (2026-10-06)
- [OpenAI Slashes API Tiers From Five to Three, Cuts Top Threshold in Half](https://alphasignal.ai/news/openai-slashes-api-tiers-from-five-to-three-cuts-top-threshold-in-half.md) (2026-10-06)
- [Ollaya Runs Decision Models Locally With Millisecond Structured Outputs](https://alphasignal.ai/news/ollaya-runs-decision-models-locally-with-millisecond-structured-outputs.md) (2026-10-06)
- [Meta Opens Rebalancer, Solving 40 Million Daily Infrastructure Problems](https://alphasignal.ai/news/meta-opens-rebalancer-solving-40-million-daily-infrastructure-problems.md) (2026-10-06)
- [Backburner Turns Your iPhone Into a Mac's AI Co-Processor, 44% Faster](https://alphasignal.ai/news/backburner-turns-your-iphone-into-a-mac-s-ai-co-processor-44-faster.md) (2026-10-05)
- [OpenBMB's MiniCPM5-2B Gets a 324M Draft Model That Accepts 6 Tokens at Once](https://alphasignal.ai/news/openbmb-s-minicpm5-2b-gets-a-324m-draft-model-that-accepts-6-tokens-at-once.md) (2026-10-03)
- [Prime Intellect Launches Prime Inference, Serving 600B Tokens Daily](https://alphasignal.ai/news/prime-intellect-launches-prime-inference-serving-600b-tokens-daily.md) (2026-10-02)
- [PyTorch and Red Hat's Helion Beats CUTLASS on H100 by 17.8%](https://alphasignal.ai/news/pytorch-and-red-hat-s-helion-beats-cutlass-on-h100-by-17-8.md) (2026-10-02)
- [Microsoft Ships WSL Containers to Replace Docker Desktop on Windows](https://alphasignal.ai/news/microsoft-ships-wsl-containers-to-replace-docker-desktop-on-windows.md) (2026-10-01)
- [Meta's TLX Beats FlashAttention-4 on Blackwell GPUs by 50%](https://alphasignal.ai/news/meta-s-tlx-beats-flashattention-4-on-blackwell-gpus-by-50.md) (2026-10-01)
- [LoopCD Cuts AI Reasoning Costs by 48% Without Changing Model Weights](https://alphasignal.ai/news/loopcd-cuts-ai-reasoning-costs-by-48-without-changing-model-weights.md) (2026-10-01)
- [Artificial Analysis' AA-AgentPerf-Local Shows RTX 5090 Beating 128GB Systems by 5x](https://alphasignal.ai/news/artificial-analysis-aa-agentperf-local-shows-rtx-5090-beating-128gb-systems-by.md) (2026-09-30)
- [Cerebras Beats Human Speed, Booking Restaurants 19x Faster Than Rivals](https://alphasignal.ai/news/cerebras-beats-human-speed-booking-restaurants-19x-faster-than-rivals.md) (2026-09-29)
- [Meta Lets CLMs Rewrite Their Own Memory, Cutting AI Compute by 35%](https://alphasignal.ai/news/meta-lets-clms-rewrite-their-own-memory-cutting-ai-compute-by-35.md) (2026-09-29)
- [Perplexity's SPACE Red Team Exposes a Network Bypass Flaw Hitting 8 Sandboxes](https://alphasignal.ai/news/perplexity-s-space-red-team-exposes-a-network-bypass-flaw-hitting-8-sandboxes.md) (2026-09-28)
- [Gimlet Cloud Adds Cerebras to Hit 3,000 Tokens per Second for AI Agents](https://alphasignal.ai/news/gimlet-cloud-adds-cerebras-to-hit-3-000-tokens-per-second-for-ai-agents.md) (2026-09-28)
- [Strata Runs a 125B AI Model on a Regular Gaming PC](https://alphasignal.ai/news/strata-runs-a-125b-ai-model-on-a-regular-gaming-pc.md) (2026-09-28)
- [Mia AI Lab Squeezes a 99 GB Qwen3.8-Flash-Next onto One GPU](https://alphasignal.ai/news/mia-ai-lab-squeezes-a-99-gb-qwen3-8-flash-next-onto-one-gpu.md) (2026-09-25)
- [Edge0 Ships Audio8 ASR Infinite to Transcribe Speech for Hours Straight](https://alphasignal.ai/news/edge0-ships-audio8-asr-infinite-to-transcribe-speech-for-hours-straight.md) (2026-09-25)
- [Liquid AI's LFM2.5-VL DSpark Makes Vision AI 3.13x Faster on Apple Silicon](https://alphasignal.ai/news/liquid-ai-s-lfm2-5-vl-dspark-makes-vision-ai-3-13x-faster-on-apple-silicon.md) (2026-09-24)
- [Databricks Buys Row Zero to Kill the Ungoverned Spreadsheet](https://alphasignal.ai/news/databricks-buys-row-zero-to-kill-the-ungoverned-spreadsheet.md) (2026-09-24)
- [Qwen3.8 Flash Next Runs a 125B Model on Dual V100s at 120 tok/s](https://alphasignal.ai/news/qwen3-8-flash-next-runs-a-125b-model-on-dual-v100s-at-120-tok-s.md) (2026-09-24)
- [Hermes Plugin Hands Agent Triage to Jev in 0.4 Seconds](https://alphasignal.ai/news/hermes-plugin-hands-agent-triage-to-jev-in-0-4-seconds.md) (2026-09-23)
- [Cohere Brings Model Vault to Canada for Enterprises Needing Data Sovereignty](https://alphasignal.ai/news/cohere-brings-model-vault-to-canada-for-enterprises-needing-data-sovereignty.md) (2026-09-23)
- [Anthropic's Opus 5.5 Cuts Agentic Coding Costs 40% While Running 30% Faster](https://alphasignal.ai/news/anthropic-s-opus-5-5-cuts-agentic-coding-costs-40-while-running-30-faster.md) (2026-09-22)
- [Amazon Brings Moonshot AI's Kimi K3 to Bedrock at 1.76x OpenRouter Prices](https://alphasignal.ai/news/amazon-brings-moonshot-ai-s-kimi-k3-to-bedrock-at-1-76x-openrouter-prices.md) (2026-09-22)
- [Meta's Petal Cable Will Move 1 Petabit per Second Across the Atlantic](https://alphasignal.ai/news/meta-s-petal-cable-will-move-1-petabit-per-second-across-the-atlantic.md) (2026-09-21)
- [Inco AI's Splash Runs Qwen3.8-27B Twice as Fast on Apple Silicon](https://alphasignal.ai/news/inco-ai-s-splash-runs-qwen3-8-27b-twice-as-fast-on-apple-silicon.md) (2026-09-21)
- [Cognition's Devin Cloud Lets Developers SSH Into AI Coding Sessions](https://alphasignal.ai/news/cognition-s-devin-cloud-lets-developers-ssh-into-ai-coding-sessions.md) (2026-09-21)
- [SGLang Squeezes 78% More AI Throughput on NVIDIA's Blackwell GPUs](https://alphasignal.ai/news/sglang-squeezes-78-more-ai-throughput-on-nvidia-s-blackwell-gpus.md) (2026-09-21)
- [MiniMax H3 Runs Locally Across Four RTX 3090s, Cutting Video Time by 2.84x](https://alphasignal.ai/news/minimax-h3-runs-locally-across-four-rtx-3090s-cutting-video-time-by-2-84x.md) (2026-09-21)
- [Ornith's DFlash Drafter Pushes Ornith-1.5-35B Decoding to 2.32× Faster](https://alphasignal.ai/news/ornith-s-dflash-drafter-pushes-ornith-1-5-35b-decoding-to-2-32-faster.md) (2026-09-20)
- [RegVGGT Lets One Consumer GPU Process Thousands of 3D Video Frames](https://alphasignal.ai/news/regvggt-lets-one-consumer-gpu-process-thousands-of-3d-video-frames.md) (2026-09-20)
- [HyperQwen Runs Qwen3.8-27B on a Single RTX 3090 at 1,035 tok/s](https://alphasignal.ai/news/hyperqwen-runs-qwen3-8-27b-on-a-single-rtx-3090-at-1-035-tok-s.md) (2026-09-19)
- [Tamara Tran's fast-jev-compaction Stops Claude Code From Forgetting Critical Tool Output](https://alphasignal.ai/news/tamara-tran-s-fast-jev-compaction-stops-claude-code-from-forgetting-critical.md) (2026-09-18)
- [Perplexity Computer Adds Four Effort Presets to Control AI Task Depth](https://alphasignal.ai/news/perplexity-computer-adds-four-effort-presets-to-control-ai-task-depth.md) (2026-09-17)
- [vLLM Boosts Kimi K3 Throughput by 2.8× With Smarter Scheduling](https://alphasignal.ai/news/vllm-boosts-kimi-k3-throughput-by-2-8-with-smarter-scheduling.md) (2026-09-17)
- [Jarrod Watts Ships Jev Trader to Make AI-Guided Crypto Trades in 100ms](https://alphasignal.ai/news/jarrod-watts-ships-jev-trader-to-make-ai-guided-crypto-trades-in-100ms.md) (2026-09-17)
- [Epoch Maps 44% of Global AI Compute Across 86 Data Centers](https://alphasignal.ai/news/epoch-maps-44-of-global-ai-compute-across-86-data-centers.md) (2026-09-16)
- [Cohere's Model Vault Now Encrypts AI Prompts Even From Cloud Admins](https://alphasignal.ai/news/cohere-s-model-vault-now-encrypts-ai-prompts-even-from-cloud-admins.md) (2026-09-16)
- [Bojie Li's Open Textbook Teaches AI Infrastructure Through Hardware Limits](https://alphasignal.ai/news/bojie-li-s-open-textbook-teaches-ai-infrastructure-through-hardware-limits.md) (2026-09-16)
- [Mia AI Lab Runs GLM-5.3-Flash Across Two DGX Sparks at 146 tok/s](https://alphasignal.ai/news/mia-ai-lab-runs-glm-5-3-flash-across-two-dgx-sparks-at-146-tok-s.md) (2026-09-15)
- [Perplexity's CobbleDB Cuts Search Storage Latency by 82% Over DynamoDB](https://alphasignal.ai/news/perplexity-s-cobbledb-cuts-search-storage-latency-by-82-over-dynamodb.md) (2026-09-15)