Skip to content
DigIntoAI/Model intelligence
Model intelligence feed

Model intelligence.

Follow the changes. Understand what comes next.

Editor’s pickGoogle
Google/Gemini 4 Neon

Google Launches Gemini 4 Neon in Antigravity With 1M Context, Four Thinking Levels, and Benchmarks vs. Claude Opus 5.5 and GPT-6.2 Sol

Announced on November 19, 2026 at 05:51 Beijing time, Google's Gemini 4 Neon features a 1M-token context window, four thinking levels, and planner-executor pairing with Gemini 4 Hydrogen in Google Antigravity, outperforming Claude Opus 5.5 and GPT-6.2 Sol across most of 18 published benchmarks.

Latest stories349

Newest published
109

OpenAI Introduces GPT-6 Cyber for Select Daybreak Red Customers

On 2026-11-06 01:09 Asia/Shanghai (2026-11-05T17:09:13Z), OpenAI introduced GPT-6 Cyber in its official blog and public post. Figures: ExploitGym 48.5% vs 32.1%, Preparedness Critical (was High), honeypot 99.93%/100% vs 94.49%. It starts via API and Codex for select approved Daybreak Red customers and will expand; turning on Daybreak Red does not grant the model. Daybreak Blue alpha for GPT-6 Astra is not Astra GA. Self-reported Linux 948 and Firefox 81; Mozilla released many fixes. We haven't applied or tested.

GPT-6 Cyber/Daybreak Red/Daybreak Blue(alpha)/GPT-6 Astra(扩覆盖非 GA)/GPT-5.6 CyberRead story
110

Gemini Omni 1.2 Flash Arrives in AI Studio and on OpenRouter

Google introduced Gemini Omni 1.2 Flash on November 6, 2026 (Beijing time), available through the Gemini API, Gemini for business, AI Studio and OpenRouter. It supports native-audio video up to 15 seconds and conversational editing. Google reports about 30% lower video prices than Omni 1.1, at roughly $0.07/second for 720p; there is no Gemini API free tier. The Gemini app rollout covers Google AI subscribers, with Ultra able to create and edit 15-second videos. Omni 1.1 requests switch to 1.2 from December 3.

Gemini Omni 1.2 Flash/Gemini API/OpenRouterRead story
111

OpenAI Private Safety Processing Reviews ZDR API Records in Your Own Cloud

At 00:54 Asia/Shanghai on November 6, 2026, OpenAI Developers said Private Safety Processing (PSP) reviews related API requests for safety without retaining prompts or responses. Records stay in the customer environment; an automated agent reviews them so personnel cannot read that content. Docs call this ZDR with PSP, writing encrypted records to AWS S3, Azure Blob, GCS, or OCI storage for 30 days. Only bounded safety signals and operational metadata leave in plaintext. The GPT-6 Cyber blog says standard retention is the default and ZDR customers generally must use PSP. We haven't tested it.

Private Safety Processing(PSP)/Zero Data Retention(ZDR)Read story
112

Cursor Status Shows Composer 2.5 Request Issues Resolved After About 50 Minutes

Cursor status opened composer-2.5 degraded (id ps_78783e1b35b98b43) at 2026-11-05 23:06:52 Asia/Shanghai (impact minor) for Composer 2.5 request issues. Overview then: Partially Degraded Service—IDE, Cursor IDE (iOS), CLI, Cloud Agents, Automations, and Bugbot degraded; cursor.com, Grok Bot, Third-party models, Origin, and Enterprise Dashboard operational. Resolved at 23:56:32 (This issue has been resolved), about 50 minutes later, no root cause. A ~00:06 Nov 6 check still shows All Systems Operational. We haven't tested; figures follow the Cursor status page and status API.

Cursor/Composer 2.5Read story
113

OpenAI Status Shows ChatGPT Work and Dots Errors—Now Resolved

OpenAI's status page/API opened ChatGPT Issues (id 01M6P50GPP1V98EF8MJ9SH393R) at 2026-11-05 23:25:03 Asia/Shanghai as investigating, impact minor, then identified at 00:57:41 and monitoring at 01:05:56 on 2026-11-06. Current name: ChatGPT Work and Dots Errors. Resolved at 12:31:37 Asia/Shanghai; status API says affected services are operating normally again and apologizes for the disruption. Overview All Systems Operational / none; unresolved=0. Not the morning paid send-message ticket resolved at 08:32:37. We haven't tested ChatGPT, Work, or Dots; figures follow the OpenAI status page/API.

ChatGPTRead story
114

OpenRouter Adds Auto Code and Moves Auto to Jev Router

OpenRouter listed openrouter/auto-code at 2026-11-05 21:12:01 Asia/Shanghai for coding agents and harnesses. It shares Auto Router's engine and picks models by coding-task pass rate, completion time, and cost per solved task. At 21:38:09, openrouter/auto's listing refreshed; the official model page now says Auto is powered by Jev Router and replaces market-spend rankings, picking model and reasoning effort from task difficulty, benchmarks, live prices, and prompt caching. Both bill at the serving model's price; context reaches 2,000,000 tokens per serving model. We haven't tested either route.

Auto Code/Auto Router/Jev RouterRead story
115

ChatGPT Dots Adds iOS Notification Replies, Sharing, and Action Button

On 2026-11-05 06:08:16 Asia/Shanghai, samir (@_samirism; Dots & ChatGPT @OpenAI, previously @Snap) said the team has been laser-focused on making dots significantly faster and more reliable, with five latest-app updates: iOS inline notification replies, share-to-dot, iPhone Action Button, faster notification tap-in, and Gmail deeplinks into the Gmail app. At 15:54:40, Tibo (@thsottiaux; Codex & ChatGPT @OpenAI) wrote “Follow Samir for dot updates”—a pointer, not an appointment. These iOS entry points have not been tapped on a real device here; Android is not mentioned.

ChatGPT Dots/iOSRead story
116

OpenAI GPT-Live Outbound SIP Calling Hits GA for All Orgs

On 2026-11-05 ~06:01 Asia/Shanghai, OpenAI’s Justin Uberti (@juberti) said the GPT-Live outbound telephony API hit GA: Live voice agents dial via any SIP trunk (e.g. Telnyx/Twilio), with HD Voice and bidirectional DTMF (Uberti’s wording). Greg Brockman (@gdb) quoted that at ~12:52. Docs: outbound SIP is for all organizations through the Live API, not the Realtime API call-creation endpoint. Ringing ≤3 minutes; connected ≤2 hours—not set in create. OpenAI’s docs and staff posts are self-reported; DigIntoAI hasn’t tested outbound calling.

GPT-Live/Live API/Outbound SIPRead story
117

Conductor Joins Meta With Cloud Wind-Down Nov 13–20 and Local Support to Early 2027

On November 4, 2026, Conductor (conductor.build) said Conductor is joining Meta (Charlie; changelog 0.93.0). New sign-ups stop that day. Conductor Cloud goes read-only on November 13 (active subscribers get a refund) and shuts down on November 20, when Cloud user-data deletion begins. Local Conductor keeps new models and critical fixes until early 2027. A newer product is only a fuzzy “early next year” promise. DigIntoAI has not tested Transfer to local, Export data, or refunds; Meta’s company blog was not separately confirmed here.

Conductor/Conductor Cloud/Local ConductorRead story
118

OpenAI to Acquire Pierre Computer Company to Strengthen Codex Git Tools

On November 4, 2026, OpenAI’s blog said it will acquire Pierre Computer Company, Inc. (“Pierre”), behind Code Storage and open-source Trees/Diffs already used to browse files and review changes in Codex. Open-source stays open; after closing, Code Storage customers can continue while OpenAI works features into Codex. Closing needs regulatory approvals; until then both stay independent. No price or close date disclosed. DigIntoAI hasn’t tested—self-reported.

Codex/Code Storage/Pierre Trees/Diffs/HighlightsRead story
119

OpenAI Daybreak Red Brings GPT-5.6 Cyber to AWS GovCloud US-East

On November 4, 2026, OpenAI’s blog said Daybreak Red—access to GPT-5.6 Cyber—is available to approved customers in AWS GovCloud (US-East) on Amazon Bedrock. High-level uses: authorized vulnerability research, exploit validation, and security testing for government security teams in controlled settings. Access needs OpenAI Daybreak Red approval and AWS provisioning; approval does not auto-enable AWS access. Not a public API; not the Paths research pipeline. DigIntoAI hasn’t applied or tested it.

Daybreak Red/GPT-5.6 Cyber/Amazon Bedrock/AWS GovCloud (US-East)Read story
120

OpenAI Cyber Pipeline Finds 948 Linux Privilege Escalation Bugs in One Run

OpenAI’s A thousand fewer paths to root (Nov 4, 2026)/@OpenAI (07:13 Asia/Shanghai) describe an autonomous cyber pipeline on an upcoming cyber model. Mid-October: 1,724 reproduced; 291 info-leak PoCs; 948 Linux LPE+patches; $1,213,430 (≈$1,279/LPE) on stock desktop Linux. Humans review; maintainers/Patch the Planet. At 09:02 Asia/Shanghai, Tibo (@thsottiaux) quoted Paths—world-class cyber team assembled; more over weeks/months. Not a public launch/pricing change; not widespread unpatched abuse. DigIntoAI hasn’t tested—self-reported.

upcoming cyber model/Patch the PlanetRead story