How I feel keeping up with all the AI news this week #GPT5 #GPT_OSS #Genie3 #Opus4
LLMS
-
GPT-5 Analysis and Key Conclusions Presentation
By
–
Antes que nada, si queréis ver todas las conclusiones sobre GPT-5, tenéis resubido el directo que acabamos de finalizar, con la presentación (interesante la primera parte) y mis análisis al principio y al final
-

GPT-5 Release: Initial Thoughts and Livestream Analysis
By
–
GPT-5 & Livestream – my initial thoughts We have been waiting a long time for GPT-5. I still remember when the first rumors about GPT-5 began to circulate, back then under the code name Gobi, later Orion, but then it was released as GPT-4.5. Now it's finally here – and
-
GPT5 Launch Brings Deep Questions About AI Future
By
–
Thanks Mark! It certainly brings up some deep questions and perhaps some clues as to the answer. Congrats on the GPT5 launch!
-

OpenAI Launches GPT-5 with Incremental Improvements and New Interface
By
–
🔴 ¡OPENAI PRESENTA GPT-5!
— Carlos Santana (@DotCSV) 7 août 2025
Con mejoras incrementales en su inteligencia, un precio muy competente, y una presentación descafeinada que no ha entusiasmado… Os cuento más detalle del nuevo modelo por defecto de ChatGPT
Bye selector de modelos! 🫡pic.twitter.com/Cto4MRZMQZOPENAI PRESENTS GPT-5! With incremental improvements in its intelligence, a highly competitive price, and a lackluster presentation that hasn't thrilled… I'll tell you more details about ChatGPT's new default model Bye model selector!
-
Grok 4 and Grok 3 Mini pricing updates compared
By
–
I updated the pricing table to add Grok 4 (same price as Claude Sonnet 4, $3/$15) and Grok 3 Mini (0.30/0.50 – compares with GPT-5 Mini which is 0.25/2.00)
-
GPT-5 ushers in intentional intelligence era
By
–
GPT-5 is a system that decides how to think. We just crossed into intentional intelligence. Welcome to the new era.
-

GPT-5 Reduces AI Sycophancy by Over 60%
By
–
And no, it’s not clingy. GPT‑5 reduced sycophancy (AI being overly flattering) by >60%. Less “you’re right” when you're wrong.
Fewer emojis.
More substance. Feels like you're talking to an actual expert not a fanboy with a CPU. -

GPT-5 Significantly Reduces Hallucinations and Errors
By
–
Less hallucinations. Less BS. Compared to GPT-4o: • 45% fewer factual errors
• 6× fewer hallucinations on open-ended facts
• 80% fewer deceptive answers on missing data GPT-5 is more accurate and more honest. -

GPT-5 Dominates Multiple AI Benchmarks
By
–
Performance? Ridiculous. • 94.6% on AIME (competition math)
• 88.4% on GPQA (PhD-level science)
• 88% on Aider Polyglot (code editing)
• 46.2% on HealthBench Hard GPT‑5 doesn’t just pass it dominates. Benchmarks look like a sweep.
