MODELS & RELEASES
Every Models & Releases story from RXed AI News over the last 30 days — 15 items.
GPT-6 Astra beat Portal in under 24 hours with zero human help after initial setup.
the-decoder.com/gpt-6-astra-beat-portal-start-to-fi…
@GoogleDeepMind's WeatherNext 3 delivers 5 km global forecasts, updated hourly, using live weather station data.
marktechpost.com/2026/09/03/google-deepminds-weathe…
@OpenAI's GPT-6 Astra is the first model capable enough to declare the AGI era, says Greg Brockman.
the-decoder.com/gpt-6-astra-is-the-first-model-maki…
Tencent's open-source Hy4 preview pairs a 770B parameter model with a 1M-token context for coding, office, gaming, and research.
testingcatalog.com/tencent-released-open-source-hy4…
Agent capability is determined more by the harness—memory, planning, tool orchestration—than the model alone.
huggingface.co/papers/2608.25593
AI agents now consume more tokens than humans on @OpenRouter, usage up 14x since February.
the-decoder.com/ai-is-becoming-ais-biggest-customer…
FlowEvo agents adapt to complex tasks by evolving workflows and skills, not discarding them after one use.
huggingface.co/papers/2607.21596
@AnthropicAI's Claude Security now uses Claude Mythos 5 to scan codebases for vulnerabilities with severity ratings and patches.
the-decoder.com/anthropic-puts-its-most-powerful-mo…
PACE-Bench evaluates AI agents adapting to changing physics environments through code evolution.
huggingface.co/papers/2608.14441
Domain data scarcity forces LLMs to repeat content, undermining token-per-parameter ratios.
huggingface.co/papers/2608.14071
How does Z.ai boost GLM-5.3 performance without retraining the 743B base model?
marktechpost.com/2026/08/14/z-ai-ships-glm-5-3-with…
@Nvidia's Nemotron 3.5 Lightning matches @OpenAI's gpt-oss-120b on the Intelligence Index with 3.6B active parameters.
the-decoder.com/nvidias-open-weight-nemotron-3-5-li…
@NVIDIA-accelerated LTX-2.5 video model generates 6.8-second clips with open weights.
marktechpost.com/2026/08/11/the-video-production-st…
Macaron-V1 open agent learns from experience and improves itself after deployment.
huggingface.co/papers/2608.09819
Pokee AI's Isaac 28B scores 93.3% on RULER at 10M tokens, customer-boundary deployment. @PokeeAI
marktechpost.com/2026/08/08/pokee-ai-releases-pokee…