The Week AI Became A Cost Question

GPT-5.6 Versus Fable

• ▸The team framed GPT-5.6 as the day's main story and planned the coverage around that comparison.
• ▸The point was not just whether 5.6 could beat Fable on raw capability. The more useful question was where it is close enough, cheaper, or easier to route into everyday work.
• ▸The agreed shape of the launch was simple: pair a video with written analysis, then make the fuller comparison available on the site.

The Workhorse Model Test

• ▸The team moved away from treating intelligence benchmarks as the whole story. A model that lands near the top model while costing far less can be the better practical choice.
• ▸Cost, reliability, and routing matter more once a model is used across lots of routine tasks instead of a few showcase prompts.
• ▸The working standard was blunt: if a model can get within a small margin of Fable at a fraction of the price, it deserves serious attention.

Open Source Pressure

• ▸The discussion treated open-source models as a growing force, especially for companies trying to keep token costs under control.
• ▸One view in the meeting was that cheaper models will handle more of the everyday workload, while the most expensive frontier models become the last stop for harder tasks.
• ▸That changes how model releases should be judged: the winner is not always the model with the highest ceiling, but the one that fits the most work.

Grok, Cursor, And The Review Cycle

Link: https://x.com/elonmusk/status/2075130137817825315?s=20

• ▸The team talked about model reviews as a recurring format, not one-off coverage, because the market is moving too fast for a single leaderboard to stay useful.
• ▸The bigger question is whether each new release gives people a clearer reason to switch, route, or rethink their stack.