Superpower Daily
Meta Adds Max Reasoning to Muse Spark 1.3 After a Two-Day Safet...Artificial Analysis Clarifies Astra’s 67-Point Coding Score Is...Anthropic’s Claude Built a 13-Million-Line Fermat Proof, Leavin...Nvidia Agrees to Buy Hugging Face for $12.93B, Promising Hardwa...Databricks Says Proteus Generated Qwen Kernels Up to 5.2× Faste...Lutnick Ties G20 AI Infrastructure Pitch to U.S. Tariff ReliefMeta Adds Max Reasoning to Muse Spark 1.3 After a Two-Day Safet...Artificial Analysis Clarifies Astra’s 67-Point Coding Score Is...Anthropic’s Claude Built a 13-Million-Line Fermat Proof, Leavin...Nvidia Agrees to Buy Hugging Face for $12.93B, Promising Hardwa...Databricks Says Proteus Generated Qwen Kernels Up to 5.2× Faste...Lutnick Ties G20 AI Infrastructure Pitch to U.S. Tariff Relief

Full archive

Latest AI News

Search reporting on AI markets, models, products, companies, policy, and culture.

Category

More filters
Clear all
Showing 1-18 of 19Results matching "vLLM"

Today

2 stories
ResearchTools

Databricks Says Proteus Generated Qwen Kernels Up to 5.2× Faster Than vLLM

Proteus specializes GPU code for particular runtime shapes rather than seeking one universal kernel. Databricks’ results highlight both the potential of that approach and the validation work needed to trust it. Databricks has described Proteus, an agent-driven system for producing specialized GPU kernels, with company-reported Qwen 3.5 122B results of 1.8–5.2× faster than the best implementations available in vLLM.

3 min readRead story
FundingStartups

Resect AI Raises $25M to Try Changing Enterprise Models Before Bad Answers Reach Users

The startup has public fact-checking models and an ambitious runtime product pitch. Its next test is showing that model-level intervention can improve reliability without disrupting useful answers. Resect AI has emerged from stealth with $25 million to build tools it says can spot hallucinations while a language model is running, then alter the model’s behavior before an unreliable answer reaches an enterprise user.

3 min readRead story

Yesterday

2 stories
LaunchModels

IFM Releases Six K2 Horizon Models With Weights, Code and Training Data

The new lineup gives developers models from 0.9B to 375B parameters under Apache 2.0, while putting IFM’s performance claims and unusually broad training record in public view. Developers seeking to move from a small local model to a larger enterprise system now have a single newly released family to test. A dense 32B model and sparse 36B-A4B model are intended for local hosting and on-premise servers.

3 min readRead story

Earlier this week

5 stories
Open releaseModels

DeepSeek Releases 305B Vision Weights Under MIT License, but Local Use Needs Serious Hardware

The release gives developers the model files and serving references needed to run or adapt DeepSeek-V4-Flash-Vision-Exp independently. Its 168GB size means that freedom is aimed more at infrastructure-equipped teams than typical individual users. The shift gives developers more control over a model that accepts images, but the 168GB checkpoint sets a high infrastructure threshold for anyone seeking to run it locally.

3 min readRead story
LaunchInfrastructure

Broadcom Validates Five AI Model Families for VMware’s On-Premises Model Service

The update gives enterprises a path to run models from NVIDIA, Google, NEC, Alibaba Cloud and Z.ai inside their own infrastructure, with new sharing, routing and code-isolation services around them. Enterprises using VMware Cloud Foundation now have a validated route to run five additional AI model families on premises and offer them internally as a model-as-a-service.

2 min readRead story

Last week

6 stories
Open releaseTools

Intel Releases 20 Agent Skills for Arc GPUs, With an Open-Port Security Catch

The open repository turns Intel’s GPU deployment steps into reusable instructions for coding agents, but several model-server launchers bind broadly without authentication by default. Intel has released 20 open-source Agent Skills that give coding agents procedures for getting AI workloads running on Arc and Arc Pro GPUs. XPU is Intel’s GPU software path in this collection.

3 min readRead story
Open releaseInfrastructure

Snowflake Open-Sources Semi-Persistence With Sub-Second Single-GPU vLLM Swaps

The release treats GPU model copies as disposable while retaining weights in CPU memory, aiming to let many specialized models share costly accelerator capacity without long reload delays. Idle models can give back GPU memory only if they can return before the next request waits too long. Avoiding the slow path on every swap The target is a tradeoff in vLLM, the inference engine Semi-Persistence currently wraps.

3 min readRead story
Open releaseModels

IBM Releases Granite 4.2 Open Weights, Reserving Agentic Training for 8B and 30B

The new family gives self-hosted deployments a choice between a smaller tool-capable model and larger variants trained for code, search and other tool-use tasks. IBM has released Granite 4.2, an open-weight language-model family built for download and self-hosting. The central distinction is not just size: its 8B and 30B models receive specialized training for tool use, while the 3B version does not.

2 min readRead story
Open releaseModels

MiniMax Maps H3 to 24GB PCs and Server Stacks, but Keeps Context Layer Hosted

The index gives developers more routes to run and adapt H3, yet the layer that turns mixed media into model-ready context remains an API product—and local use is restricted in several major markets. MiniMax has published an H3 integrations index that points developers from a 24GB local setup to ComfyUI workflows, Apple Silicon, fine-tuning, and multi-GPU serving through SGLang and vLLM-Omni.

3 min readRead story
BenchmarkInfrastructure

NVIDIA Says Vera Rubin Delivers 30x More Agentic Throughput per Megawatt, Pending Review

The preliminary comparison shifts attention from fixed prompt benchmarks toward the power required to keep long, tool-using AI sessions responsive. NVIDIA says Vera Rubin NVL72 delivered up to 30 times the throughput per megawatt of GB300 NVL72 for an agentic AI workload at 160 output tokens per second per user. NVIDIA reports the DeepSeek V4 Pro 1.6T comparison on the AgentX workload.

3 min readRead story

August 2026

3 stories
BenchmarkModels

Open Models Are Catching the Frontier Faster. Benchmark Scores Aren’t the Whole Contest.

SemiAnalysis sees the lag from closed-model breakthroughs to open-model benchmark parity shrinking across three AI eras. But a comparable model score still leaves a harder product question: who can turn capability into dependable agent work? In the reasoning and agentic eras, open AI models reached closed-model benchmark parity in months, according to SemiAnalysis ’s cross-era analysis.

4 min readRead story
1 more next