In August 2025 OpenAI released GPT-5 powering ChatGPT, claiming fewer factual errors, stronger coding ability, and new agent-like capabilities such as calendar access and email drafting. Anthropic rolled out Claude 4 in May 2025 and Claude 4.1 in August 2025, delivering extremely large context windows (hundreds of thousands of tokens) and adding Claude Code for…
Read more
GPT-5 launched on August 7, 2025 and was rolled out to all ChatGPT users, including the free tier. OpenAI CEO Sam Altman described GPT-5 as “clearly generally intelligent” and a “PhD-level” expert, a step toward AGI but not AGI yet. GPT-5 features “Think Harder” mode with test-time compute that automatically allocates more processing power for…
Read more
In late 2024, MiniMax released Hailuo’s first text-to-video model “Video-01,” and its current Hailuo-02 can generate 6‑second 720p videos at ~25 fps from text prompts and images. As of January 2025, Hailuo added a multi-shot “Director Mode” for controlled shot composition and an audio-generation feature to add sound or speech to videos. December 2024 saw…
Read more
Grok is Elon Musk’s xAI chatbot built on the Grok LLM family, with Grok-4 released in July 2025 and Grok-3 trained in a 200,000-GPU data center codenamed Colossus, plus built-in web search by late 2024 and multimodal image generation via Aurora. Grok-4 is integrated into Tesla vehicles as an onboard voice assistant as of July…
Read more
WIZ.AI is a Singapore-based voice-centric conversational AI provider that launched TalkGPT in 2023 as the region’s first ChatGPT-powered customer engagement solution. WIZ.AI claims 98% of users cannot tell they are speaking with a bot due to hyper-localized vocabularies and the ability to mimic local accents. By 2024, WIZ.AI had 300 clients across 17 countries, expanding…
Read more
MIT and Duke researchers used AI to design polymers with iron-based ferrocenes that absorb stress, dramatically speeding up additive discovery from weeks per candidate, with findings published in ACS Central Science. Profluent Bio used a protein language model trained on 500 million sequences to invent OpenCRISPR-1, the first CRISPR enzyme designed entirely by AI, which…
Read more
NVIDIA Drive Thor delivers up to 1,000 TOPS (INT8 sparse) on a single chip and can be paired via NVLink-C2C to about 2,000 TOPS for high-end Level 4–5 autonomy. Thor combines a Blackwell GPU with 2560 CUDA cores and 96 Tensor Cores, a 14-core Arm Neoverse V3 CPU complex, on a 4nm TSMC process, and…
Read more
Qualcomm Snapdragon X Elite (Oryon) is built on a 4 nm process, houses 12 high‑performance Oryon cores up to 4.3 GHz, an Adreno GPU around 4.6 TFLOPs, and a Hexagon NPU up to 45 TOPS with dual micro NPUs in an always‑on sensing hub. The X Elite includes dual micro NPUs in an always‑on Sensing…
Read more
NVIDIA’s Blackwell B200 features 180 GB of HBM3e memory per GPU with up to 8 TB/s bandwidth, 18 PFLOPS FP4 tensor throughput, 9 PFLOPS FP8, and 4.5 PFLOPS FP16, plus a second-generation Transformer Engine. NVIDIA claims DGX B200 delivers about 3× the training performance and 15× the inference performance of DGX H100 in end-to-end workflows.…
Read more
Sora can generate videos up to 20 seconds at 1080p for ChatGPT Pro users and 10 seconds at 720p for ChatGPT Plus, with a research version capable of about 1 minute. Sora outputs at 24 fps and supports widescreen, vertical, and square aspect ratios, with visible watermarks by default. Sora excels at photorealistic imagery and…
Read more
