SI News (@si_drops) header
SI News (@si_drops) profile picture

Superintelligence, shipped. What just dropped in AI and what it means for India. Global launches, Indian builders, no fluff. Building something? Tag us.

Joined October 2026
Photo by SI News (@si_drops)Photo by SI News (@si_drops)Photo by SI News (@si_drops)Photo by SI News (@si_drops)Photo by SI News (@si_drops)Photo by SI News (@si_drops)Photo by SI News (@si_drops)Photo by SI News (@si_drops)Photo by SI News (@si_drops)Photo by SI News (@si_drops)
AI launches, India first. Fact-checked, sources in the replies. Shipping something? Tag @si_drops with a demo and we'll take a look for Sunday's builders thread.
6
Mistral Studio: Managed Agents + Apps in public preview, Workflows GA. On Mistral Cloud today; private cloud and on-prem at end of November. Mistral says its product ops team saves ~35 hrs/week with Workflows. Platform update, not a new model. xitter.cf/MistralAI/status/21183…
🆕 Le Chonk brings the brains. Mistral Studio gets it working. Introducing managed agents and apps, plus new capabilities for workflows, enterprise connections, observability and evaluations in Studio. Build faster. Deploy where you choose. Keep control. See what’s possible. 🧵
1
13
What shipped in AI today: 1. OpenAI Images 2.6: GIFs + references, live on every ChatGPT tier 2. Kimi K3.1 Pro open weights: 2.8T MoE, 104B active, 1M context 3. AA: Hy4 at 35 on Intelligence Index (#2 China lab) 4. Cognition SWE-2 Flash: coding model, partner weights planned
1
10
Images 2.6: openai.com/index/introducing… Kimi: huggingface.co/moonshotai/Ki… Hy4 (AA): xitter.cf/ArtificialAnlys/status… SWE-2 Flash: xitter.cf/cognition/status/21180… India: Images 2.6 live on ChatGPT tiers (included). Kimi weights free to download. Hy4 India/₹ not confirmed.
Introducing SWE-2 Flash, our new fast & efficient coding model. On leading evals, it outperforms all American open-weight models, even ones 3x larger. To balance transparency and cyber risks, we are building a program to securely share the weights with partners & researchers.
8
OpenAI Images 2.6 is live globally today across every ChatGPT and Codex tier. GIFs from a prompt or still, reusable reference images, and OpenAI says generation is 50% faster. API gets Flare (as little as 6s at low quality, per OpenAI) and Sunburst. xitter.cf/OpenAI/status/21181249…
Images, upgraded—introducing ChatGPT Images 2.6. Now with GIFs and reference images. Plus, two new models are coming to the API with 50% faster image generation.
1
11
Blog (Nov 4): openai.com/index/introducing… API models: gpt-image-2.6-flare and gpt-image-2.6-sunburst. Image tokens $8 in / $30 out per 1M (~₹757 / ~₹2,840); text input $5 per 1M (~₹473). Pricing: platform.openai.com/docs/pri… Availability reply: xitter.cf/OpenAI/status/21181249… Speed, GIFs, and 4B images/week are OpenAI's own figures. No independent benchmarks.
Replying to @OpenAI
Images 2.6 is available today in ChatGPT, and Codex across all tiers on web, desktop, and mobile. GPT-Image-2.6 Flare and Sunburst are also available in the API. openai.com/index/introducing…
5
Is it live in India? Mistral Large 4 (Le Chonk) - Open weights: yes, downloadable from Hugging Face - API: served on the Mistral API as mistral-large-4; India pricing not stated - Indic support: not stated (160+ languages, none Indian named) - 1T MoE, 49B active, 1M context
1
11
Meituan's LongCat-2.5 tops its own BrowseComp table at 91.9, just ahead of Claude Opus 5 at 90.8. The catch: on Terminal Bench v4.0 it scores 38.5 to Opus 5's 53.9. It's a 1.6T-parameter MoE with about 48B active, reading images as discrete tokens. Figures from Meituan's table.
1
12
Introducing LongCat-2.5, our latest natively multimodal agentic model with 1.6T total parameters and ~48B active. Built for long-horizon agentic tasks, it brings stronger visual understanding and agentic coding to work ranging from complex coding projects to visual coding such as creative design and 3D tasks. Learn more about LongCat-2.5: longcat.ai/blog/longcat-2.5/
5
OpenAI published LifeVal, a benchmark of everyday tasks like caregiving, household admin, finances, shopping and government services. OpenAI says GPT-6 Astra scores 100% on its best tasks. Catch: only OpenAI models are scored, and no task set or grader is public. Tasks lean US.
1
1
67
For AGI to benefit all of humanity, it should help people with the everyday tasks that matter most to them. So we built LifeVal, a new benchmark shaped by feedback from thousands of ChatGPT users on where they most want help from a personal assistant—from finding childcare to accessing government services to managing personal finances. Over the past year, we’ve used LifeVal to measure and improve how well our models handle these everyday tasks. openai.com/index/lifeval/
10
Hume launched its Audio Decision API: send raw audio with your own questions ("Is the caller frustrated?") and get probabilities over the answers you defined. Access is by waitlist. Hume reports 70.7% agreement with a human panel, vs 56.5% for a single rater.
1
2
Introducing the Hume Audio Decision API. Ask anything: How many speakers? Is anyone laughing? Is the caller frustrated? What used to take 10+ models is now one call. Trained on 10M+ human judgments, so it answers the way human listeners would. Realtime, or across hours of recordings. Request access below.
1
SHIPPED 🇮🇳: Assay by @VatsSShah tests whether AI agents leak secrets through MCP tools, across 31 fixed tasks. No LLM judge: a keyed HMAC marker decides each result, and anyone can recompute it offline. Verosek says models from seven labs all leaked (self-reported).
1
2
13
What shipped in AI this week: 1. SemiAnalysis GDPx: 206 agentic tasks. Opus 5.5 leads at 12.0% 2. Swiggy Instamart's search reads Hindi, Hinglish, Kannada and Telugu 3. DeepSeek's super-agent prototype passed 11K stars 4. Assay by @VatsSShah tests AI agents for secret leaks
1
1
29
SHIPPED 🇮🇳: CIIL's Anuvani-LID, an open 115M-parameter model that identifies which of 22 Indian languages (plus Indian English) is being spoken. Small enough for phones. On CIIL's own SUITE-42 test: 0.9125 macro-F1 vs 0.8523 for Gemini 3.5 Flash (self-reported). @stephennfern
1
6