More models, More problems 💰

New model releases and why new models feel worse somehow

• GLM-5.3: Frontier Coding with Emergent Cyber Capabilities
  Z.ai released the newest refresh of their GLM 5 model. It’s still text only but promises frontier coding abilities similar to Fable.

• Introducing Gemini 3.7 Flash
  Google also released a new model even though 3.6 Flash was just released a few weeks ago. It promises increased capabilities that are comparable to Sonnet 5.

• Introducing Toast 1
  Mixedbread released a search agent that should save up to 10x the tokens with better result quality.

• Agentic Code Quality
  Addy Osmani thinks that constraints and rigorous testing are the way forward as human code review doesn’t scale.

• Understanding is the new bottleneck
  The author makes the argument that understanding has to stay with humans because agents become better at verifying their own results.

• Why does Opus 5 feel worse to work with?
  If you have also felt Opus 5 to be a downgrade in working with this article is for you.

• Maximizing the value of your Claude Code sessions
  A good guide on how to save tokens and getting the most out of your limits.