Ali Çevik (@alihcevik) profile picture

Gemini API Product Lead at Google DeepMind Y-Combinator Alum (S14)

Joined December 2009
Photo by Ali Çevik (@alihcevik)Photo by Ali Çevik (@alihcevik)Photo by Ali Çevik (@alihcevik)Photo by Ali Çevik (@alihcevik)Photo by Ali Çevik (@alihcevik)Photo by Ali Çevik (@alihcevik)
Big milestone today. We're opening up Genie 4 to developers, and adding WebSocket and WebRTC support to the Interactions API via a new Live Interactions protocol. Early step towards our vision of Interactions API becoming a universal interface to intelligence, which is inherently multimodal and live.
Shipping Genie 4, now available to developers! You can try it in the Gemini API and AI Studio. Generate worlds at 24fps with native audio, based on text or image, and then steer it live with WASD
6
3
29
890
API team shipped almost daily this week, but this one is real special :)
Introducing Gemini 4 Hydrogen, an extremely capable and fast model built to bring Gemini 4 to everyone! Available to all tiers of Google AI Studio, Gemini API, Gemini for business, the Gemini app, Google Search, and paid Google AI Pro + Ultra subscribers in Google Antigravity.
2
20
718
Ali Çevik retweeted
Gemini Managed Agents can now run in your own Daytona sandboxes. Gemini drives the agent, and every task runs in its own isolated computer that you control. Get started: daytona.io/docs/en/guides/ge…
building with Managed Agents on the Gemini API? now you can bring your own sandbox with self-hosted environments start building today: ai.studio
3
7
3
38
2,350
And second API launch of the day. You can now use managed agents with your own compute.
Self-hosted sandboxes are live in Managed Agents Connect any environment and the Gemini Managed Agent will offload tools to run inside compute you control.
1
15
648
We've seen decisions emerge as a really useful primitive in the ecosystem over the past few months for a wide range of use cases where full generation is overkill. That led us to build the Gemini Decisions API, bringing fast, calibrated decisions across text, images, audio, and video. Still early and iterating fast. Happy building and let us know what we should fix! aistudio.google.com/docs/dec…
Today we’re launching the Gemini Decisions API, powered by our new Gemini Decision 1 model in @GoogleAIStudio. Instead of generating text token by token, it’s built for closed-set choices scoring up to 1,000 questions across text, images, audio, video + PDFs in a single pass. A few details: - 3 decision types: yes/no, multiple choice (up to 255 options) + rubric scoring - $0.10 / 1M input tokens with $0 billed for output tokens (~5x cheaper than Flash-Lite) - leads on several benchmark evals with lowest calibration error Google teams are already building with it, below is a look at Stitch Rapid designing a full product flow in seconds (1x speed)
4
1
14
507
We launched a thin wrapper over Gemini API you can give to your existing agent setup (e.g antigravity, openclaw, Claude code etc) that enables generating images, audio, text with your AI Studio API key, without your agent having to write curl commands. Give it a go!
We launched the Gemini API CLI. A CLI for calling the Gemini API. Give it to the agent you already use (Claude Code, Codex, Antigravity, Cursor, OpenClaw, Hermes...). It's NOT a new coding agent you can use it too: 1. Generate images, video, speech, music. 2. Analyze PDFs, images, audio, video, YouTube URLs, 3. Transcribe recordings to md / json / srt, 4. Run Deep Research in the background. Stateless. No memory, no chat mode. Your agent owns the context and the plan, gemini-api does the API call. Build for Agents, it's self-describing. `--help`, `--dry-run`... Start here: aistudio.google.com/learn/in…
16
3
119
9,084
All the primitives are there, point your coding agent to build your own personal coding or knowledge work tool using Gemini API.
Because the Gemini API exposes the Antigravity harness + remote Linux sandbox as simple primitives, you can trivially build your own custom Antigravity Web Studio in a single Python file. Check out Example #5 in `gemini-agy-examples` github.com/clementfarabet/ge…, where a simple web UI lets you interact with the remote Antigravity instance, and control sandboxes.
15
702
Quality of live improvement on interactions api with safety settings.
shipped safety_settings on the Interactions API today so you can block jailbreak attempts and filter unsafe inputs/outputs right in the API call instead of running a separate moderation pass. aistudio.google.com/docs/saf…
3
240
Ali Çevik retweeted
Gemini 4 Argon (gemini-4-argon) is now live across Google AI Studio, the Gemini API, Google Antigravity, Gemini Enterprise, and the Gemini app! Gemini 4 Argon is awesome: 1. Unified 1M Context & 1M Output (Long Decode): Input, thinking, and output now share a single 1M-token window—lifting the old 64K output cap by 16x. With background=True on the Interactions API, our servers handle multi-hop continuation and checkpoint recovery automatically so the model can think for tens of minutes and emit complete multi-file repo migrations or formal proofs in a single pass. 2. Tunable Reasoning Depth: Control compute per turn via thinking_level ("low" | "medium" default | "high") plus native agentic video processing (media_processing="agentic"). 3. Real-World Engineering: 77.9% on DeepSWE v1.1, 91.9% on Vibe Code Bench, and 68.9% on the Vals Index. Internally, we used Argon to migrate 800K+ lines of Fuchsia Zircon C/C++ kernel code to memory-safe Rust and replace 32K lines of hand-written SIMD in libgav1 for a 2.7x speedup. 4. Token Economics: Launching at an introductory rate of $2 / 1M input and $10 / 1M output tokens through Nov 30, with 95% off cached inputs ($0.10 / 1M) via automatic implicit caching.
Gemini 4 Argon is here. It’s available globally today for developers and enterprises in the paid tier of the Gemini API and the Gemini platform, for Ultra subscribers in Google Antigravity, and Ultra and Pro subscribers in the Gemini app and Search. Hope you all enjoy it!
11
12
129
6,323
Now that personal agents are finally working, people will want to interact with most software/services agentically. For app developers this means one of two paths: 1. integrate with personal agents (MCP server, CLI) 2. turn the app itself into an agent (hosted agent/harness) On (1), integration: - goal is to make your app reachable by personal agents, open up a new distribution channel - but this means giving up direct touch with end users as more of them interact this way, which makes you swappable - also harder to differentiate on user experience. the app is driven through a straw (the MCP server or CLI) into someone else's context window/filesystem, and you have little control over the end to end journey - you can't really encapsulate business logic: your skill/plugins live on the client side, you can't properly eval them or keep your secret sauce - and your app can't really invoke the personal agent on its own, so the relationship isn't bidirectional - but it's cheap to own and run, you just maintain few endpoints and the user's own agent subscription pays for the tokens On (2), becoming an agent: - you remain in control of the experience: the app owns the loop rather than narrating every step to a caller. business logic (e.g skills) stays on your servers, you can control and eval them properly - your app can be proactive, watching its own signals and starting conversations when something changes, the way Openclaw works in the background - it can show up where users already are (Slack, Google Docs, SMS, email), remember them across sessions, and answer from what it knows itself rather than from whatever fits in someone else's context window - but you pay for the tokens, and you maintain a third major surface after web/desktop and mobile - for now distribution is for you to figure out, although this will likely get better as agent to agent interactions get solved What is not clear yet is whether end users will want one agent that does everything, or app specific agents. The single agent shares context across apps and is simpler, but likely weaker on any individual journey. Historically individual apps have been preferred for high value services.
6
1
15
1,435
Alongside our new family of sota Gemini 3.8 TTS models, we shipped a new Voices API which allows you to design voices. Create reusable assets using voices api, then use interactions api for generating speech. aistudio.google.com/docs/voi…
8
4
51
2,777
Continuing to ship: 1. New harness version that performs better 2. New Files API that allows you to upload/download files from your live agent sandbox 3. Credentials API, support for oauth, without your agents seeing your credentials
We're updating Gemini managed agents with a new harness. antigravity-preview-09-2026 brings @Antigravity tools and behavior to @GoogleAIStudio and the Interactions API. It runs natively on Gemini 3.8 Flash, our best reasoning & coding model yet, and you can change your model selection anytime. Plus, we’re introducing two new APIs to make building easier and more secure 🧵
8
1
2
77
12,022
We take developer documentation seriously. 1. @timeyoutakeit from Vercel joined Gemini API team earlier this year, leading all our documentation efforts. 2. Her starter project was: rebuild our docs platform. The existing platform (ai.google.dev/gemini-api/doc…) served us fine for the past few years, but it never felt like a native docs platform. Now @timeyoutakeit has rebuilt it natively. These two will co-exist for short period of time for compatibility but we'll move everything over eventually. As usual, send over feedback :) Lots to improve still.
today we’re sharing the first preview of the Gemini API docs right in AI Studio we built the experience from the ground up to give you better consolidation and convenience: docs right where you’re building get started today: ai.studio/docs
9
3
1
51
9,218
We do take the feedback under this thread quite seriously. 👇
what's your early feedback on Gemini 3.8 Flash? where should we improve?
47
2
1
119
35,107
We've just launched agentic video processing on the Gemini API. Instead of loading the entire clip into context upfront, you can mark video inputs for agentic processing so the model pulls in parts dynamically as needed. This cuts token usage and improves response quality on longer videos. If you’ve been trimming or pre-processing videos to save context, this should give you better results since the model is trained specifically for this workflow. Try passing the full video unedited, mark it for agentic processing and let me know how it performs. ai.google.dev/gemini-api/doc…
2
3
66
3,115
It feels like we still haven't figured out how to use AI for thinking and this bothers me. Since I read this essay by PG couple years ago, I've been cautious. paulgraham.com/writes.html TL;DR writing == thinking so if you're not the one doing the writing, you're also not the one doing the thinking. However it is clear that the answer can't be don't use AI to write or think. I've been experimenting with a dedicated thinking skill but curious what others are doing?
9
2
35
3,692
We've just opened up access to new Gemini Omni 1.1 Flash model on the Gemini API. There are bunch of improvements here (e.g draft mode, upscaling) but what I'm most excited about is the ability to extend an existing video. It uses 10s of prior context from an existing video to extend it. API docs : ai.google.dev/gemini-api/doc…
Stop generating entire AI videos from scratch just to change one detail. Gemini Omni 1.1 Flash is here on the Gemini API & @GoogleAIStudio - we worked on giving full creative control to developers. What you can build with now: 🎬 Extend scenes keeping 10s of context and character lock 🎯 First and Last frame conditioning ⚡ 360p draft mode - low cost, lightning fast ✨ 1080p and 4K upscaling 📹 Video references 💬 And of course, iterative editing using natural language conversation
9
2
1
37
4,527
Yesterday, we added Gemini 3.5 Transcribe to the Gemini API. While you could prompt regular LLMs for speech-to-text, this model is purpose built: faster, enables accuracy for transcription (e.g verbatim transcripts without unintended edits and hallucinations, word-level timestamps). It also supports removing filler words (e.g "ums", "ahs") and custom vocabulary. ai.google.dev/gemini-api/doc…
Introducing Gemini 3.5 Transcribe, our most precise speech-to-text model yet, designed for intelligent voice interactions! 💬 Available for both real-time speech and audio file transcription, with support for: 💡 smart transcription: rewrite and format text based on intent. 🌍 support for 85+ languages, regional accents and diverse dialects. 📚 custom vocabulary: recognize jargon and specific spellings. 💯 high precision in alphanumerics. 🫂 multi-speaker identification.
2
3
31
3,149
Another week, another great set of improvements to managed agents. My personal favourite is hooks - you can now define scripts that run within the environment before or after certain tool calls. @tryoffdeal uses post tool hooks to verify company logos produced by the managed agent are accurate and on-brand.
1
1
1
17
1,556