GPT-5 arrives with 1.8T parameters and multimodal native training

OpenAI's next-generation flagship model is trained natively on text, image, audio, and video from day one — eliminating the patchwork architecture of prior generations. Early benchmarks suggest 34% improvement on multimodal reasoning tasks.

Sep 27, 2026 · Featured

Read full coverage

Latest Releases

Llama

Llama 4 70B released with MoE architecture

Sep 26, 2026

Open SourceLLM

Meta's latest open-weight model uses mixture-of-experts to deliver frontier-class performance while running on a single A100 GPU.

Gemini

Google

Gemini 2.5 Pro adds real-time video understanding

Sep 25, 2026

Closed SourceVision

Google's flagship multimodal model now processes live video streams with sub-100ms latency for real-time analysis.

Qwen

Alibaba

Qwen 2.5-VL 72B outperforms GPT-4V on DocVQA

Sep 24, 2026

Open SourceMulti-modal

Alibaba's open-source vision-language model sets a new benchmark for document understanding, surpassing closed-source competitors.

Claude

Anthropic

Claude 4 Opus ships with 1M token context

Sep 23, 2026

Closed SourceLLM

Anthropic's new flagship model handles a million tokens of context, enabling full-book and entire-codebase reasoning.

Whisper

OpenAI

Whisper V4 adds 50 more languages

Sep 22, 2026

Open SourceAudio

OpenAI's speech-to-text model now supports over 100 languages with significantly improved accuracy on low-resource languages.

Sora

OpenAI

Sora 2 generates 10-second videos at 60fps

Sep 21, 2026

Closed SourceMulti-modal

OpenAI's video generation model doubles frame rate and introduces consistent character animation across shots.