Ivan Solovyev

Product manager @ Google DeepMind

[Ivan Solovyev @solovyev_i]

@HeyGen HyperFrames

[Ivan Solovyev @solovyev_i]

learn more in the blog: https://t.co/lHwWOdl4kG

[Ivan Solovyev @solovyev_i]

Meet Gemini 3 Flash ⚡️

Frontier intelligence built for speed and scale.

🚀 3x faster than 2.5 Pro 🧠 90.4% on GPQA Diamond 👀 Code execution for visual inputs 💸 $0.50 / 1M input tokens

Available to build with now in Google AI Studio & Gemini API.

[Ivan Solovyev @solovyev_i]

We've just released a new Gemini Native Audio model!
Expect improved function calling, better instruction following and more accurate context retrieval.

Give it a go at: https://t.co/CTvdQnppDx

More in the blog: https://t.co/37b4ArxafK

[Ivan Solovyev @solovyev_i]

@siculars @OfficialLoganK Gemini-2.5-pro-preview-tts is distinct from Chirp 3. It is a Gemini-based model that provides much more flexibility and control over the output, resulting in significantly better performance than Chirp.

[Ivan Solovyev @solovyev_i]

More in the blog post: https://t.co/Nkb6fd9Q6i

[Ivan Solovyev @solovyev_i]

New Gemini 2.5 Flash and Pro TTS models are here!

We focused heavily on giving you more control: richer expressivity, context-aware pacing, and much better multi-speaker dialogue.

You can start vibe coding apps with these new capabilities right now in Google AI Studio.

[Ivan Solovyev @solovyev_i]

@vigneshmani1512 @OfficialLoganK @_philschmid Embeddings are generated automatically. It's just 3 steps: create a store, upload your files, and send your query. The system handles chunking, indexing, and creating embeddings for you.

[Ivan Solovyev @solovyev_i]

@Siyaxn @OfficialLoganK @_philschmid Yes! Just upload the files you want to index, and you can start querying.

Make sure to update the documents when your code changes to keep the results fresh.

[Ivan Solovyev @solovyev_i]

@sreeenidhi Thanks for flagging this! You can create as many stores as you want. The only limit is on the total storage space you use, which depends on your quota tier. The docs are now updated to reflect this.

[Ivan Solovyev @solovyev_i]

@hey_liot @OfficialLoganK @_philschmid Thanks for flagging this! You can create as many stores as you want. The only limit is on the total storage space you use, which depends on your quota tier. Updating the docs now.

[Ivan Solovyev @solovyev_i]

@mrlnonai AI Studio App https://t.co/05iZyibkRg is a good frontend example, you can use it as a code sample or remix into a different experience.

[Ivan Solovyev @solovyev_i]

We’ve just launched FileSearch, our integrated RAG solution for Gemini API. Just upload your files and start querying.
The best part? You only pay for embeddings during indexing. Storage is completely free.
Details: https://t.co/8nnECjAVXL

[Google AI Studio @GoogleAIStudio]

https://t.co/MfFF7AgE0K

[Ivan Solovyev @solovyev_i]

@richardt830 @OfficialLoganK @_philschmid Vertex RAG Engine is designed for high customization, letting you build your exact pipeline.

FileSearch focuses on simplicity. We handle the management tasks so you can go straight from uploading a file to your first query.

[Ivan Solovyev @solovyev_i]

@judeandryan Concurrent session limit is removed for Gemini 2.5 Flash Preview Native Audio. Gemini 2.5 Flash Live and Gemini 2.0 Flash Live limits have not changed.

[Ivan Solovyev @solovyev_i]

Good news for Live API developers: we increased token rate limits for native audio model and removed the concurrent session limits. https://t.co/Ak2lZeudtf

Happy building : )

[Ivan Solovyev @solovyev_i]

@mesMntainG2 @OfficialLoganK Savings from serving optimizations, like implicit caching, are built directly into the Live API's per-token price. While the Live API doesn't currently have direct support for implicit cache, you'll still benefit from these improvements through lower overall costs.

[Ivan Solovyev @solovyev_i]

We heard your reports about the latest gemini-2.5-flash-native-audio-preview-09-2025 model cutting off mid-response. We've recently rolled out a fix to solve this.

The model should be much more stable now. Please keep the feedback coming!

[Ivan Solovyev @solovyev_i]

You can now enable thinking for the latest Gemini Live model! 🧠

Give the Live model a "thinking budget" to unlock deeper reasoning for your most complex questions. It’s a powerful new way to get more thorough answers.

Try it now in AI Studio: https://t.co/iiM1C7sVLB

[Ivan Solovyev @solovyev_i]

@OfficialLoganK @ahmtaydn_9 You can tell Gemini how fast to speak directly in your prompt or system instructions. For example, try:
"Talk really fast at a rate of 30 words per second". Models perception of time is not perfect yet, so you may need to crank that WPS rate higher to get desired speed.