This is in part "abuse prevention"[1] and in part marketing. Making customers feel like they're signing up to access state secrets makes the models seem more "special". Sama is well known to use these SV marketing tricks, like invite-only access, waiting lists, etc to psychologically manipulate users into thinking they're begging for entry to an exclusive club instead of just swiping a credit card to access an API.
Google tried this with Google Plus and Google Wave, failed spectacularly, and have ironically stopped with this idiotic "marketing by blocking potential users". I can access Gemini Pro 2.5 without providing a blood sample or signing parchment in triplicate.
[1] Not really though, because a significant percentage of OpenAI's revenue is from spammers and bulk-generation of SOE-optimised garbage. Those are valued customers!
Gemini doesn't give you reasoning via API though, at least as far as I'm aware.
If by reasoning you mean showing CoT, Gemini and OA are the same in this regard - neither provides it, not through the UI nor through the API. The "summaries" both provide have zero value and should be treated as non-existent.
Anthropic exposes reasoning, which has become a big reason to use them for reasoning tasks over the other two despite their pricing. Rather ironic when the other two have been pushing reasoning much harder.
Google exposes their reasoning. You can use their new gemini python sdk to get thought traces.
Google does not expose their reasoning any more. They give "thought summaries" which provide effectively zero value. [1][2]
[1] https://ai.google.dev/gemini-api/docs/thinking#summaries [2] https://discuss.ai.google.dev/t/massive-regression-detailed-...
Works for me?
Maybe you’re thinking of deep research mode which is web UI only for now.