pandaily.com /1 week ago

UniPat AI's SPU Benchmark Asks AI Agents to Predict Real Wet-Lab Results, and None Score Above 0.5

UniPat AI's SPU benchmark has AI agents rank the outcomes of 4,207 real wet-lab experiments. The best of 13 configurations averaged 0.486 out of 1.

Read source
pandaily.com /1 month ago

StartLux-Decision v1.1 Tops Decision Index 0.3.2 With Open Models That Think Only When Unsure

StartLux's 27B Instant model leads the community-run Decision Index 0.3.2 with a Full score of 69.12. The family spans eight open models from 0.8B to 27B under a non-commercial lic...

Read source
geeky-gadgets.com /1 month ago

RTX 3090 Beats 64GB Mac Studio with 140 TPS Local AI Speed

Choosing the right hardware for running local AI models like Qwen 3.6, a 35-billion-parameter system, involves balancing performance, memory capacity and cost. The Stack explores t...

Read source
pandaily.com /1 month ago

Axera Unveils AX8860 Edge AI Chip Series With 256 TOPS and Up to 128 GB Memory

At CIIF 2026, Axera launched the 6nm AX8860 for on-site large-model inference, with AXLink cascading, PCIe 5.0 and a vLLM/SGLang-friendly toolchain.

Read source
editorialge.com /2 days ago

GPT-6.2 Astra vs Claude Fable 5.2 vs Gemini 4 Carbon: What the AI Race Really Measures

GPT-6.2 Astra scores 62, Claude Fable 5.2 65 and Gemini Carbon 60. Explore the benchmark limits, earlier screenshots and Chinese competition. The post GPT-6.2 Astra vs Claude Fable...

Read source
pandaily.com /1 month ago

Simate Debuts Simate-beta Physical AI Fast System, Says It Tops RoboDojo Benchmark

Three-month-old Simate released Simate-beta, a fast-reaction physical AI model it says ranks first on RoboDojo with a 33.95 average score, developed with an agent-driven research p...

Read source
pandaily.com /1 month ago

Alibaba's 27B XekRung Model Tops CyberGym Model Leaderboard at 88.9%

Alibaba Security AGI Lab's XekRung-1.5-27B-Preview, fine-tuned from Qwen3.8-27B, reached an 88.9% success rate on UC Berkeley's CyberGym Level 1 and now ranks first on the benchmar...

Read source
dev.to /1 month ago

Four big local LLMs on one RTX 5090 and 128 GB of RAM, benchmarked side by side

One RTX 5090, 128 GB of RAM, and four open models from 125B to 748B parameters. Over the past weeks I got each of them running on the same workstation, on three different runtimes....

Read source
techmeme.com /4 weeks ago

Artificial Analysis says Kimi K3.1 Pro lands 4th on its Intelligence Index, behind only Opus 5.5 (max), Sonnet 5.5 (max)...

@artificialanlys: Artificial Analysis says Kimi K3.1 Pro lands 4th on its Intelligence Index, behind only Opus 5.5 (max), Sonnet 5.5 (max), and Muse Wonder Full (max)  —  Kimi K3.1...

Read source
pandaily.com /3 weeks ago

XHEART Launches X7 Ultra Driving Chip With 700 TOPS, Targets 3,000 TOPS With X9 in 2027

XHEART's X7 Ultra brings 700 TOPS for end-to-end driving models and a production Armv9 automotive CPU. Its next chip, the X9, targets 3,000 TOPS and 700GB/s of memory bandwidth in...

Read source
pandaily.com /1 month ago

XPeng and Banma Run a 30B-Class Cockpit AI Model On-Device on XPeng's 750-TOPS Turing Chip

XPeng and Banma Intelligence built a cockpit solution that runs a 30B-class AutoOmni model locally on XPeng's 750-TOPS Turing chip, keeping core AI features working offline.

Read source
arxivgpt.medium.com /2 weeks ago

Darwin-180B-RSI Ranks #1 Open-Weight Model in Math and Science Reasoning on SuperCLUE (Nov 2026)

Continue reading on Medium »

Read source
geeky-gadgets.com /3 weeks ago

Deepseek V4.1 Flash Hits 1,286 Tokens per Second in New Tests

Deepseek V4.1 Flash has emerged as a standout in the AI landscape, blending speed, versatility and cost-efficiency to tackle a wide range of challenges. As highlighted by Matthew B...

Read source
dev.to /1 month ago

A 180B model that tops seven Hugging Face leaderboards now runs on a laptop — no GPU needed

Running a frontier-class model used to mean a rack of data-center GPUs. Not anymore. VIDRAFT has released POCKET-Darwin-180B, a compressed build of Darwin-180B-RSI — the model tha...

Read source
pandaily.com /1 month ago

Dalian University of Technology's VA-Bench: Top Multimodal Models Finish Only Half of Robot Tasks

VA-Bench from Dalian University of Technology tests 12 multimodal model setups on robot-arm tasks; Qwen3.8-max leads at 53.93%, and none finishes a strict long-horizon task.

Read source
cnx-software.com /3 weeks ago

$2600 LattePanda Sigma 2 single board server features 180 TOPS Intel Core Ultra X7 358H SoC for local AI applications

The LattePanda Sigma 2 is a compact single board server/computer powered by an Intel Core Ultra X7 358H Panther Lake-H processor with Intel Arc B390 graphics and a 50 TOPS NPU deli...

Read source
theregister.com /3 days ago

To rank the world's biggest supercomputers, we're gonna need a bigger bench

AI has birthed some truly colossal supers, they just aren't on the Top500

Read source
testmuai.com /1 month ago

11 Best AI Evaluation Tools for Production in 2026

The 11 best AI evaluation tools for production in 2026, ranked by seven criteria checked in vendor docs: online scoring, CI gates, alerts and self-hosting.

Read source
pandaily.com /1 week ago

Goke Plans GK6950 Edge AI Chip Series for Embodied Robots, With 10 TOPS and VLA Support

Goke plans the GK6950 edge AI chip series for next year: 10 TOPS INT8, mixed precision for interactive models and FP16 spatial reconstruction.

Read source
pandaily.com /3 weeks ago

AgiBot's GO-3 Nano Preview Tops RoboDojo's Simulation Leaderboard With a 61.90 Average Score

GO-3 Nano Preview leads the RoboDojo simulation leaderboard with an average score of 61.90 and a 56.42 percent success rate, ahead of second place by 6.64 points.

Read source
nextbigfuture.com /1 month ago

Top AI Models Every Few Days

Google is releasing Gemini 4 Argon and it is competitive. There is Claude Sonnet 5.5. It is far cheaper to use. Use it on high. New models are being released almost every day. Clau...

Read source
cryptobriefing.com /2 weeks ago

DeepSeek V4.1 Flash hits 848 tokens per second on Nvidia’s GB300 NVL72

Rapid software advancements in AI models can significantly alter capacity planning and cost structures, impacting scalability strategies. The post DeepSeek V4.1 Flash hits 848 toke...

Read source
pandaily.com /1 month ago

Robotera's VPP2 World Action Model Tops RoboDojo, Scores 58.5% Zero-Shot on Real ALOHA Arms

Robotera's open-source VPP2 tops RoboDojo's 42-task dual-arm benchmark at a 32.26% success rate by training video prediction first and robot actions second.

Read source
blog.google /5 days ago

New AI training performance benchmarks for Google TPUs

In the latest MLPerf Training v6.1 benchmark round, Google Cloud published 10 results across seventh-generation Ironwood TPUs and NVIDIA Grace Blackwell GPUs, spanning f…

Read source

Turn fresh research into a full content calendar

Use SocialBu to discover ideas, generate post drafts, and schedule them across your social channels.