AYUSH SAINI

engineer building

[AYUSH SAINI @ayushsainime]

@grok @MassViper66 @grok who do you work for ?

[AYUSH SAINI @ayushsainime]

API credits: OpenAI now includes monthly API credits with eligible ChatGPT Pro and Business plans, from @OpenAIDevs (OpenAI).

Per OpenAI, eligible subscribers can claim up to $500 per month to build apps and agents. Pro plans get credits equal to the plan price ($100, $200 or $500). Business seats get $20 (Standard) or $100 (Premium), pooled and capped at $500 a month. To claim, the paid plan must be at least 7 days old and linked to one API organization you own, with the same email on both accounts. Plus and free trials don't qualify, credits expire at the end of the billing cycle, and they don't cover Codex used through API keys.

Where to read it:
1. https://t.co/dLgQpS7k8O
2. https://t.co/yFHkOYRS6V

Follow Ayush Saini for more such content

#OpenAI #ChatGPT

[AYUSH SAINI @ayushsainime]

Cloud earnings: Alphabet Q3 and Microsoft FY27 Q1 results, from @sundarpichai (Google) and @Microsoft.

Per Alphabet, Google Cloud revenue grew 103% to $30.7B, with TPU system sales a contributor. Per Microsoft, Azure grew 48% to $33.2B, the first time Microsoft gave that figure in dollars (per GeekWire). Microsoft's CFO Amy Hood, as quoted by GeekWire, said "Customer demand continues to exceed available supply." Company figures, not advice.

Where to read it:
1. https://t.co/Qx04u2OxuN
2. https://t.co/q2GdHhkqSa
3. https://t.co/QwlgzEHO4q

Follow Ayush Saini for more such content

#GoogleCloud #Azure #AI

[AYUSH SAINI @ayushsainime]

Vercel CI: a new beta CI service that runs only the tasks a change touches, from @vercel (Vercel).

CI is the automatic check that runs when code is pushed. Per Vercel, time from push to passing checks inside Vercel got 2 to 6x faster, with nearly 8 million tasks run so far. It is in beta on all plans, Linux and GitHub repos only, and Pro is priced at $0.0035 per CPU minute per Vercel's docs. Agents can read CI logs through vc ci commands or Vercel MCP. The speed figure is Vercel's own.

Where to read it:
1. https://t.co/oiI6OsXoDE
2. https://t.co/9E5lYdjAPw
3. https://t.co/JVB1ZEbWuq

Follow Ayush Saini for more such content

#VercelCI #Turborepo #AgenticCI

[AYUSH SAINI @ayushsainime]

ViBench 2.0: a benchmark that asks AI agents to build a whole app, from Replit.

A benchmark is a test that scores AI. Per Replit, this one has 17 products and 241 feature requests, and GPT-6.1 Sol leads with 65% of full builds passed (about $12 a build), ahead of GPT-6 Astra at 60% (about $90) and Opus 5.5 at 49%. On the 8 public products, per Replit, GPT models went from 0% (GPT-5.4) to 31% (GPT-6 Sol) to 53% (GPT-6.1 Sol). Per Replit, nearly three in four failed builds fail only where features meet. Read with care: Replit sells an app-building agent, one model family (Opus 5.5) grades, and the validation numbers are Replit's own.

Where to read it:
1. Replit blog
https://t.co/kZUhY4IWCo
2. Public repo (8 products)
https://t.co/sczCvwxdQF

Follow Ayush Saini for more such content

#VibeCoding #CodingAgents #AIBenchmarks

[AYUSH SAINI @ayushsainime]

Ayoooo..@NVIDIAAI liked this post..

[AYUSH SAINI @ayushsainime]

Nemotron 3.5 Transcribe: an open-weight speech model, from @NVIDIAAI (NVIDIA).

One model transcribes 40 languages, translates 31 languages to and from English, and labels who said what in 13 languages, on recordings up to 1 hour with up to 8 speakers. Per NVIDIA, it transcribes 100 hours of audio in under 3 minutes on one RTX PRO 6000 (NVFP4) and ranks first among open-weight models on the Open ASR Leaderboard. Those benchmarks are self-reported. About 4.3B active parameters, OpenMDW-1.1 license (commercial use allowed). It is offline only, with no live streaming. Hindi is rated "Leading" on the model card. NVIDIA also released Granary v2, an open speech dataset of 463,062 hours in 48 languages.

How to try it:
1. Weights on Hugging Face (BF16, FP8 and NVFP4):
https://t.co/m6hnhrzr5T
2. NVIDIA's blog with benchmarks and Granary v2:
https://t.co/y1uu5o8tdI
3. Serving recipe for SGLang-Omni:
https://t.co/Y7eu8PgOzm

Follow Ayush Saini for more such content

#VoiceAI #OpenSource

[AYUSH SAINI @ayushsainime]

Android MCP: Google's way for Android apps to offer their tools to AI assistants like Gemini, from @AndroidDev (Android).

Google announced it at Android Dev Summit. MCP is a standard way for apps to offer tools to AI assistants. Per Google, the AppFunctions API (Android MCP) is in beta, and the Connected Apps Platform is a preview with phased access, so users could ask Gemini to use an app. Google says 50+ Connected Apps have launched, including Adobe, Experian, Localiza and Peloton in Gemini. Who gets access, where, and any revenue or ranking details are not disclosed.

Where to read it:
1. https://t.co/UWJof5za2w
2. https://t.co/aqq5dhtCEj
3. https://t.co/x3kUBe4aLA

Follow Ayush Saini for more such content

#AndroidMCP #MCP #AndroidDevSummit

[AYUSH SAINI @ayushsainime]

Copilot Fusion: GitHub's AI coding helper that picks the best AI model for each job, now open to all Copilot plans, per GitHub, from @github.

Per Artificial Analysis, it scored 60 against 62 for Claude Code with Opus 5.5 Max, at $5.59 against $11.42 per task. Per GitHub's Octoverse, over 40% of pull requests (proposed code changes) now start with an AI draft, up from 17% in March. Microsoft claims up to 77% lower cost, which is its own number.

Where to read it:
1. https://t.co/MxDE9Ty1qO
2. https://t.co/D17tIRZyIT

#GitHubUniverse #CopilotFusion #AICoding

[AYUSH SAINI @ayushsainime]

Particle-1: a new AI model for writing code that aims to use less text for the same results, from @Suhail (Particle).

Per Particle's own tests, Particle-1 uses about 45% fewer tokens (small pieces of text the AI reads or writes) than DeepSeek V4.1 Flash, the model it is built on. Particle reports its pass rate (share of coding tasks solved) at 72.57% vs 71.15%, within the margin of error, and cost per task 34% to 52% lower. These are Particle's own numbers. It is on Particle's API, not a free download.

Where to read it:
1. https://t.co/zdezBJ1eeR
2. https://t.co/YgIqrXVD99

Follow Ayush Saini for more such content

#Particle1 #CodingAgents #TokenEfficiency

[AYUSH SAINI @ayushsainime]

7 latest updates by grok bot @bot https://t.co/ZzcbFBEIAH

[AYUSH SAINI @ayushsainime]

here we go again ..

[Google AI @GoogleAI]

Announcing Gemini 4 Carbon, our strongest model with state-of-the-art cybersecurity and long horizon capabilities.

Carbon represents a step up across most benchmarks, delivering strong frontier performance in autonomous coding and cyber defense.

We’re first making Carbon available to a subset of trusted cyber defenders in our Fairwind program. We’ll expand access as soon as possible.

[AYUSH SAINI @ayushsainime]

Responses API: a new API for building agents on Mistral, in public preview, from @MistralAI (Mistral AI).

Per Mistral, each response returns the model's answer, its reasoning and its tool calls as separate items, and it runs on all Mistral-hosted chat models, including Mistral Large 4. It follows the OpenAI Responses format and is stateless: your app keeps the history and resends the output items. The docs list text, image and file inputs, JSON mode and JSON Schema output, streaming and client-executed function calling. It is a public preview, so expect changes.

How to try it:
1. Install the SDK:
pip install -U mistralai
2. Call the endpoint:
POST /v1/responses
3. Read the guide:
https://t.co/gXd09RJ1Nd
4. API reference:
https://t.co/kTCSFWCHzN

Follow Ayush Saini for more such content

#AIAgents #MistralAI

[AYUSH SAINI @ayushsainime]

Claude for Legal: a gallery of ready-made legal artifacts, from @claudeai (Anthropic).

Per Anthropic, the gallery has six artifacts it built itself, including Cite check, Regulatory horizon scan and Tabular review, plus four built by Harvey, Goodwin, PwC and Foley & Lardner. Each one opens with an example, and teams add their own documents and connectors. Anthropic says it has added more than 60 legal connectors since Claude for Legal launched. Anthropic also lists nine access-to-justice artifacts, five built by nonprofits it supported with Claude credits.

Where to read it:
1. Anthropic's announcement:
https://t.co/WTrw5iIwl6
2. The launch post on X:
https://t.co/NEZW5XoVi8

Follow Ayush Saini for more such content

#ClaudeForLegal #LegalTech #LegalAI

[Claude @claudeai]

New in Claude for Legal: a gallery of ready-made artifacts for contract review, regulatory horizon scans, cite checks, and more.

Start from the built-in example, add your own documents and connectors, and share it with anyone across the business.

[AYUSH SAINI @ayushsainime]

Slack Code API: bring your own coding agent into Slack, from @SlackHQ (Slack).

It is now generally available on all Slack plans. A coding agent opens its own code channel, adds people and files, posts diffs, HTML previews and its thinking steps, and your team steers it right there. Agents inherit the channel's permissions. Launch companies include Anthropic, Cognition, GitHub and Vercel. Per Slack (self-reported), GitHub Copilot already opens a code channel in 30% of its agent sessions, and 150+ customers and others asked for access. Merging PRs from Slack is on the roadmap, not shipped yet.

Where to read it:
1. Slack's announcement:
https://t.co/WpiSYrBN4p
2. Developer docs (agents.conversations.* and agents.sessions.*):
https://t.co/evmSzqq5ik

Follow Ayush Saini for more such content

#AIAgents #CodingAgents

[AYUSH SAINI @ayushsainime]

Gemini 4 Carbon: Google's new frontier model for long coding jobs and cyber defense, from @GoogleDeepMind (Google). Announced today, two weeks after Gemini 4 Argon.

Per Google, Carbon is #1 on the Artificial Analysis Intelligence Index v5.0 at 59 and tops 15 of the 17 benchmarks in its launch table, against Argon, GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1. Cyber scores: SEC-bench Pro 88.2%, ExploitGym 54.8%, CWE-bench 71.0% (run without guardrails). It still trails Opus 5.5 on OSWorld-2.1 and GPT-6 Astra on FrontierSWE 2.

You cannot use it yet. A group of cyber defenders gets it first, then trusted partners through the Fairwind Program. Developers, businesses and consumers come later, with no date given. Intro price: $2 per 1M input tokens, $10 per 1M output, rising to $4 and $20 later.

Where to read it:
Google's launch post
https://t.co/aDx3bMi73i
Fairwind Program
https://t.co/F9UoVVXv1Q

Follow Ayush Saini for more such content

[AYUSH SAINI @ayushsainime]

Genie 4: Google DeepMind's new world model for real-time interactive worlds, from @GoogleDeepMind (Google).

Per Google, you steer a running world with text prompts (move the camera, change the weather, add characters), and every world generates its own audio, including dialogue. It is a research preview for trusted testers today. Google says it is coming soon, starting with Google AI Ultra subscribers in Project Genie (from $99.99 a month in the US). Project Genie runs Genie 3 today. Google says a limited group of early-access developers in the Gemini API and Google AI Studio gets it soon too. Sessions run a few minutes for now.

Also new this week: SpAItial's Echo-3 turns iPhone captures of a real space into a 3D scene at real-world scale.

How to try it:
1. Join the Genie 4 waitlist from Google DeepMind's page:
https://t.co/34JlxgE5Ox
2. Read the announcement and see the limits:
https://t.co/ZdirUo6mJp
3. Check the Google AI Ultra plans:
https://t.co/AHoqS9E2Pn
4. Echo-3, iOS app now, Android coming soon:
https://t.co/QrqKrMX6jT

Follow Ayush Saini for more such content

#Genie4 #GoogleDeepMind #WorldModels

[AYUSH SAINI @ayushsainime]

iCloud Mail and iCloud Calendar in ChatGPT: two new plugins, from @OpenAI (OpenAI).

Per OpenAI, iCloud users can now connect Mail and Calendar to ChatGPT to find emails, catch up on their inbox, check their schedule and update calendar events. It is available on all ChatGPT plans on the web, the desktop app and iOS (workspace app controls still apply). Each plugin connects on its own with an Apple app-specific password, never your regular Apple password.

How to try it:
1. Make an app-specific password. Sign in at
https://t.co/ukIhKDNLe2
then Sign-In and Security, App-Specific Passwords, Generate an app-specific password. Your Apple Account needs two-factor authentication.
2. In ChatGPT, open Settings > Plugins and install iCloud Mail.
3. Sign in with your iCloud Mail address (@icloud.com, @me.com or @mac.com) and the app-specific password. Type it only in the plugin's connection form, never in the chat.
4. Repeat for iCloud Calendar, this time with your primary Apple Account email.

Apple's guide to app-specific passwords:
https://t.co/q6Sy5evv6H

OpenAI's setup notes:
https://t.co/W30X1kMKrd

Follow Ayush Saini for more such content

#ChatGPT #iCloud #OpenAI

[AYUSH SAINI @ayushsainime]

@hailports @AnthropicAI Yea , i think The real breakthrough isn’t just speeding up discovery, but making it reproducible. If independent labs can consistently replicate the results, that’s when AI-driven science gets really exciting.

[AYUSH SAINI @ayushsainime]

Mithril Medicine: the one-person biotech Anthropic just acquired, plus early results from six drug developers using Claude, from @AnthropicAI (Anthropic).

Per Anthropic, founder Cody Tranbarger used Claude to coordinate 11 outside labs and took the work from an idea to promising results in mice in under 18 months. Anthropic says the work is mostly preclinical: human trials could start as soon as late 2028, and only if the drug candidates continue to hold up. The acquisition price was not disclosed.

Where to read it:
1. Anthropic's research post:
https://t.co/OGB9TGQhMZ
2. Anthropic's announcement on X:
https://t.co/blXiCL40ad

#Claude #Anthropic #AIforScience

[Anthropic @AnthropicAI]

Over the past few months, Anthropic scientists have engaged in deep research partnerships with six organizations developing new medicines. We've also begun drug discovery efforts of our own.

In doing so, we’ve learned much more about how Claude can support the many stages of drug development.

In this blog post, we share some early results from these efforts, and describe how this work fits in with our wider ambitions in the life sciences.

https://t.co/EVbxkdDTgJ

[AYUSH SAINI @ayushsainime]

@daisyineurope Jensen huang