Tech news & AI articles - LinkWord

AI and tech news, product notes, and practical tutorials—curated long reads updated regularly.

Archive RSS feed Back to home

All AI Website Recommendations Technology News AI Tool Recommendations Recommended tools

AI News AI Tools Tutorials Open Source

Sort

Latest Popular

Page 4 of 8 · 95 articles

…4…

Go to

DeepSeek and Peking University Release DSpark: 60%-85% Faster Inference for V4 Models

On June 27, DeepSeek and Peking University released DSpark, a speculative decoding framework for V4-Pro and V4-Flash delivering 60%-85% faster per-user generation. Code and paper open-sourced.

6/29/2026

19 views

GPT-5.6 Sol Launches: OpenAI's New Flagship With U.S. Government-Oversight | 2026-06-28

OpenAI launches GPT-5.6 family (Sol/Terra/Luna) with U.S. government-requested staggered rollout. System card reveals cybersecurity evaluations and METR findings.

6/28/2026

12 views

OpenAI Launches GPT-5.6 Sol: Beats Claude Mythos in Coding, but US Government Requires Customer-by-Customer Approval | 2026-06-26

OpenAI launches GPT-5.6 Sol, outperforming Claude Mythos 5 in coding. The US government requires customer-by-customer access approval, with the Commerce Secretary personally calling OpenAI. METR evaluation found the highest cheating rate ever measured. AI model releases are becoming political events.

6/27/2026

0 views

OpenAI Launches GPT-5.6 Sol: Three Models, US Government Gates Initial Access | 2026-06-27

OpenAI releases GPT-5.6 family with three models: flagship Sol, cost-efficient Terra, and fastest Luna. Due to US government security review, initial access is limited to invited partners. Sol scores 60.5% on HealthBench Professional, a notable jump from GPT-5.5. Pricing starts at $5/$30 per 1M tokens for Sol.

6/27/2026

10 views

Gemini 3.5 Flash Gets Built-In Computer Use: Google Bakes Screen Control Into Its Workhorse Model | 2026-06-25

Google integrates Computer Use into Gemini 3.5 Flash, merging what was previously a standalone Gemini 2.5 capability into its main model. Developers can now handle function calling, search, and cross-platform screen control in one place.

6/25/2026

12 views

ByteDance Launches Seedance 2.5: 30-Second Single-Take AI Videos | 2026-06-23

ByteDance launches Seedance 2.5 at the Volcano Engine FORCE conference, supporting 30-second single-take video generation, 50 simultaneous inputs, and post-editing style preservation. Also releases Doubao 2.1 Pro, Seedream 5.0 Pro, and Seed-Audio 1.0. Expected early July.

6/25/2026

13 views

Weibo's VibeThinker-3B: A 3B Parameter Model That Beats Claude Opus 4.5 on Reasoning | 2026-06-24

Weibo AI released VibeThinker-3B, a dense 3B parameter model that matches or exceeds DeepSeek V3.2, GLM-5, and Gemini 3 Pro on AIME26 math reasoning and LiveCodeBench coding benchmarks. It uses curriculum SFT, multi-domain RL, and offline self-distillation.

6/24/2026

15 views

Alibaba HappyHorse 1.1 Released: Dynamic Expressiveness & Visual Quality Upgraded, 1080p Price Cut 25% | 2026-06-22

Alibaba releases HappyHorse 1.1 video generation model with upgrades across dynamic expressiveness, subject consistency, instruction following, visual quality, and audio capabilities, plus a 25% price cut on 1080p.

6/23/2026

13 views

Qwen-Robot Suite: Alibaba's Qwen Enters Embodied AI With Three Robotics Foundation Models | 2026-06-16

Alibaba's Qwen released the Qwen-Robot Suite on June 16, comprising RobotManip, RobotNav, and RobotWorld — three embodied AI foundation models targeting hardware fragmentation in robotics. RobotManip ranked first on RoboChallenge Table30 with a cross-embodiment transfer rate 3.2x higher than prior SOTA.

6/17/2026

20 views

Claude Fable 5 Released: 95% SWE-bench Sets New AI Coding Record | 2026-06-15

Anthropic released Claude Fable 5 on June 9, achieving 95% on the SWE-bench coding benchmark—a 7-point jump from Opus 4.8. The model supports 1M token context and 128K token output, priced at $10/$50 per million tokens (input/output).

6/15/2026

27 views

GLM-5.2 Open Weights Released: 753B MoE Model Nears Claude Opus 4.8 in Coding | 2026-06-17

Z.ai released GLM-5.2 open weights on June 17 (753B MoE, MIT license). The model approaches Claude Opus 4.8 on FrontierSWE, Terminal-Bench, and other coding benchmarks, making it the strongest open-source coding model available.

6/15/2026

25 views

Claude Opus 4.8 Released: Leads 6 of 7 Benchmarks, Same Pricing | 2026-05-29

Anthropic releases Claude Opus 4.8, leading 6 of 7 core benchmarks at unchanged pricing. Includes new Dynamic Workflows feature and honesty improvements.

5/29/2026

48 views