Just to give an example, I ingest otel trace spans individually and in a materialized view calculate the total duration of the whole trace among other things.
"Modern cell phones, we observed a dozen years ago, are
“such a pervasive and insistent part of daily life that the
proverbial visitor from Mars might conclude they were an
important feature of human anatomy.” Riley v. California,
573 U. S. 373, 385 (2014). Since then, the percentage of
Americans who own smartphones has only increased. To-
day, more than nine in ten Americans own a smartphone.
See W. Bishop, Pew Research Center, Mobile Fact Sheet
(Nov. 20, 2025) (91%); compare A. Smith, Pew Research
Center, Smartphone Ownership—2013 Update (June 5,
2013) (56%)."
Using gpt-5.4-mini in off-peak hours already feels like super-speed to me. That's probably no more than 100-150 tk/s. I can't imagine 750!
I've always eyed Cerebras but never had a use for it that would justify paying for the API directly. Although now that I think about it, trying out the API would probably cost less than a subscription for a month...
Easily the most interesting part of this announcement is buried in the second to last paragraph:
"We're also launching GPT‑5.6 Sol on Cerebras at up to 750 tokens per second in July, bringing frontier intelligence to customers at unprecedented speed. Access will initially be limited to select customers as we expand capacity."
750 tokens/s on a frontier model is going to be extremely interesting. I doubt this new version is anything but a version bump in terms of capabilities but if we can start getting these answers back faster, they end up being more useful.
Just off the top of my head, I can think of the tedious task of finding certain functionality within a codebase. I usually can't beat an AI agent harness at this task today. If the AI model is 3x faster I have less of chance.
I use both the openai subscription and the opencode go subscription. I use the go subscription for my personal work and the openai subscription for my consulting work.
The differences between the models are minimal, but I usually stick with gpt-5.4-mini, gpt-5.4, mimo-pro-2.5, deepseek-v4-pro. These latter ones have way more usage than even using 5.4-mini so I tend to use them in personal projects for that reason.
"Death of Silicon Valley" in this case is such a funny perspective. Like, how twisted is the US's view of the market that they think "Competition? Oh no. Sound the alarms."
It's the only model where an explicit instruction at the end of my message is sometimes ignored. This doesn't happen with any of the gpts, kimis, glms, qwen, etc. Just a deepseek problem.
I use it through my opencode go subscription and it's exactly how you described. Very pragmatic and not too ambitious. It's similar to Kimi 2.5/6 in that regard.
Have you tried DeepSeek V4 Flash? It's very competent and extremely cheap.
I think Gemma 4 is also a good example of a capable small model.
I mention these not only because they're cheap but because they can run on consumer devices. The "every year bigger and more capable SOTA model" trend is mirrored by "the every year smaller and more capable open source model" trend.
But...AWS is a platform too, no? Seems like you're in the same category of risk you just moved to a more well-known name. Granted, Amazon is the most reliable even if they have their own quirks.
And backups. Sqlite makes it easier but no backup process is easy. You always have to backup and restore at least once to have the confidence to rely on it.
It's another (big) point towards paying someone else to host it.
Just to give an example, I ingest otel trace spans individually and in a materialized view calculate the total duration of the whole trace among other things.