I’m on vacation, but the AI arms race never stops.
Two bits of news that might interest you.
- GPT 6 dropped. There was nowhere NEAR as much hype around this compared to both GPT-4 and 5, but apparently this latest improvement is insanely good. I don’t have access, yet, but the benchmarking results and early reviews seem very promising. Just as well, as I have a new personal project I’m keen to get started on, so it has come at the perfect moment. Crazy to think that three months ago people thought the OpenAI team were completely washed…
- Muse Spark 1.3 is here. A total sleeper launch, and one that only insiders will know about or care. Meta, formerly known as Facebook, formerly known as The Facebook, has been building their own dream team of AI talent, only to pump out the least impressive, borderline embarrassing models. Muse Spark 1.3 is their first serious effort that qualifies as, perhaps, frontier. While perhaps not as good as Fable, Opus, Sol, and now Astra (the GPT 6 name), it’s 95% there, and 95% cheaper. On the contributor tier, it’s $0.10 output, $0.20 input per million tokens. Compare that to Fable and Sol, which are $10 and $40 respectively. Granted, at the contributor tier you’re sharing your info with the model, but some (many?) people don’t care, and the guarded version is still only a dollar or two per million tokens. I don’t know what game Zuckerberg is playing, but it’s cool knowing there is a great, super affordable agent out there that isn’t Chinese.