The first GPT-6 model

The new Astra model, potentially the first of the GPT-6 family, shows impressive capabilities in tasks like image generation and complex simulations, though it is described as ‘token-hungry’ and not consistently great. Despite initial token consumption issues and the user’s own ‘skill issue’ perception, demos showcase significant potential in areas like 3D world building and autonomous UI generation. OpenAI has also achieved its ‘automated research intern’ goal, with further advancements targeting automated AI researchers.

• Astra, the first GPT-6 model, is powerful but consumes a large number of tokens.
• Users are creating advanced applications like 3D worlds, games, and simulating real-world objects.
• Astra demonstrates capabilities in image generation, complex simulations, and potentially autonomous research.
• OpenAI has achieved its automated research intern goal and aims for automated AI researchers by March 2028.
• Anthropic is testing ways to extend Claude Code’s functionality with plugins.
• Inworld has developed Realtime TTS-2 for low-latency voice generation in AI applications.
  https://bender.layer3.press/articles/8fa8007b-ecdf-4b9d-8f2e-92f1a0a9e9ab