Rohan Taneja

✨ I made a tiny macOS app to help prevent digital eye strain by following the 20-20-20 rule! Every 20 minutes, it covers your screen for 20 seconds, forcing you to give your eyes a break and glance away from your laptop. Let me know what you think 🙂 apps.apple.com/us/app/glance-…

Interfaze is now live on Vercel's AI Gateway ▲ vercel.com/ai-gateway/mod…

@robinebers @OpenRouter @vercel @vercel_dev @aisdk added! vercel.com/ai-gateway/mod… vercel.com/ai-gateway/mod… vercel.com/ai-gateway/mod…

why is Seedream on @OpenRouter but not @vercel AI Gateway? @vercel_dev @aisdk

@sean_snd @hp_arora @joshuawolk

Hey @rtaneja_ @hp_arora why is the search on ai gateway synced between Browse Models and API Keys?

@hp_arora @tomhacks @vercel and @ZYPX4 🫡

@tomhacks @vercel nice work @rtaneja_

Yo @vercel... what?!

@theCTO @theo branchfurniture.com

i need a new office chair, recommendations? i bet @theo has at least one

You can now tell AI Gateway to fail over before a provider's default timeout kicks in. Set a custom timeout per-provider, with 𝚙𝚛𝚘𝚟𝚒𝚍𝚎𝚛𝚃𝚒𝚖𝚎𝚘𝚞𝚝𝚜, for more granular control. In beta for BYOK, with non-BYOK support coming soon. vercel.com/changelog/prov…

The world's fastest Flux Schnell provided is now on Vercel's AI Gateway. From Prodia.

Grok Imagine Video and Grok Imagine Image are now on AI Gateway. Use them for free through February 25th.

Grok Imagine Video is now on AI Gateway. Text-to-video. Image-to-video. Video editing. Audio. 𝚊𝚠𝚊𝚒𝚝 𝚐𝚎𝚗𝚎𝚛𝚊𝚝𝚎𝚅𝚒𝚍𝚎𝚘({ 𝚖𝚘𝚍𝚎𝚕: '𝚡𝚊𝚒/𝚐𝚛𝚘𝚔-𝚒𝚖𝚊𝚐𝚒𝚗𝚎-𝚟𝚒𝚍𝚎𝚘', 𝚙𝚛𝚘𝚖𝚙𝚝: '𝚋𝚒𝚛𝚍𝚜 𝚊𝚌𝚛𝚘𝚜𝚜 𝚜𝚞𝚗𝚜𝚎𝚝' });

@ram_01_ram @Yuvi_1011001 @techwraith @ZYPX4 @vercel_dev @vercel_support @aisdk we're working on it - I'll post an update here when it's ready!

@rtaneja_ @Yuvi_1011001 @techwraith @ZYPX4 @vercel_dev @vercel_support @aisdk Any updates on fix for same timestamp on model created.

@techwraith @ZYPX4 @vercel_dev @vercel_support @aisdk Need fix on Response API. In request payload, "input": [ { "type": "message", "role": "user", "content": "Hello" } ] "type": "message" seems to mandatory field, but openrouter have this as optional, and openai doest have.

@Yuvi_1011001 @ram_01_ram @techwraith @ZYPX4 @vercel_dev @vercel_support @aisdk thanks for the feedback, will take these back to the team!

@ram_01_ram @rtaneja_ @techwraith @ZYPX4 @vercel_dev @vercel_support @aisdk @ZYPX4 In model list dashboard, please add the created date as well, also make the list sorted based on newly created model. This is useful to get quick glance on recent model availability.

@ram_01_ram @techwraith @ZYPX4 @vercel_dev @vercel_support @aisdk fixed! "type": "message" is now optional

@techwraith @ZYPX4 @vercel_dev @vercel_support @aisdk Can we make "type": "message" optional in the Response API? Many apps follow OpenSpec, where it isn’t required. Because of this, I can’t use Vercel AI Gateway directly in app where existing payloads don’t include "type". The same app works fine with OpenRouter.

AI Gateway provides day 0 support for the OpenResponses API. • Use one API for multi-provider AI calls • Route requests across providers via AI Gateway • Switch providers by changing the 𝚖𝚘𝚍𝚎𝚕 string

You can now access our models via @vercel's AI SDK and AI Gateway 🥳 vercel.com/ai-gateway/mod…

@i_am_brennan @hp_arora forwarding to the team and we’ll take a look!

@rtaneja_ @hp_arora I have a lot of random failures, but I'm not doing a good job of tracking them. TBH i fully expected these to be visible within the vercel dashboard. Grok returns forbidden randomly like maybe for a few hours every day. These are evals so literally the same inputs.

@rtaneja_ @hp_arora Does the AI Gateway not use responses endpoints with the reasoning models? Per openai it's better.

@kinder_grinder @shaper cool! will take this back to the team

Thanks for listening. I am not sure how the tech implementation should be done, but indeed, video generation takes time. In my app, I am using AI SDK to dynamically generate steps for agents and tools. I have image-gen agent using Nano Banana, and I wanna expand to video generation and different models while I still stay with AI Gateway.

Hey friends @shaper @rtaneja_ any plans on adding Sora 2 support in AI Gateway?

@i_am_brennan @hp_arora could you give more info about what the warning/error is? what does the request you're sending look like? are you by any chance setting both temperature and topP?

@rtaneja_ @hp_arora Oh, wait what?! There’s another issue with anthropic models on bedrock (I ended up pinning to avoid). Bedrock hard errors on a validation where Anthropic warns. Leads to random failures… but is that a gateway problem or bedrock?

@i_am_brennan @hp_arora fixed! thanks for reporting, and lmk if you have any other questions or still see the error

@rtaneja_ @hp_arora I get errors, when it routes to the azure randomly, from azure that say something like: this model requires the use of the responses api and you’ve used the chatcompletion. Codex model was the one most frequently erroring. This was today

@i_am_brennan @ryancarson x.com/rtaneja_/statu…

@ryancarson This is good news at least x.com/rtaneja_/statu…

I think the Gemini team is going all-in on the Instructions API. If the model labs diverge on their primary generation API interface, it’s going to be bad times for all of us building on top of LLMs. I can’t get locked into a one-lab solution. It’s never going to be good for my customers. Curious how @aisdk is going to handle this.