#5 · Models · GoogleThu, Nov 19, 2026
Google rolls out Gemini 4 Neon
Google released Gemini 4 Neon, positioned between Gemini 4 Hydrogen and Carbon, with a 1M-token context window and four selectable thinking levels, priced at $3 per million input tokens and $15 per million output tokens. It replaces Argon in the Gemini app as Gemini 3.1 Pro is retired, rolling out through the API, Google’s business offering, and paid consumer products. The launch extends Google’s rapid Gemini 4 rollout (Carbon launched October 29) and sharpens the three-way frontier race with OpenAI and Anthropic.
Key people/companies: Google; Google DeepMind.
Models · OpenAINov 17–18, 2026Benchmarks vendor-reported; “High” rating is OpenAI’s own assessment
OpenAI releases GPT-6.2 Astra and Sol — with a rocky reception from early testers
OpenAI launched GPT-6.2 Astra, its frontier model for complex professional work and scientific research, and GPT-6.2 Sol, its efficient everyday model, available in ChatGPT Work, Codex, and the API. Astra claims record scores on Agents’ Last Exam at under a third of GPT-6 Astra’s cost per task, with API prices cut 20% ($8/$40 per million input/output tokens). But several early testers publicly criticized Astra’s autonomy and usability — saying it asked permission too often and struggled with autonomous workflows. Notably, GPT-6.2 Astra is reportedly the first externally released OpenAI model rated “High” for AI self-improvement capability.
Caveat: benchmark claims are vendor-reported, and the “High” self-improvement rating is OpenAI’s own assessment under its own framework.
Key people/companies: OpenAI.
Models · Google DeepMindThu, Nov 19, 2026
Google opens Genie 4 interactive world model in the Gemini API
A day after its Project Genie launch, Google made Genie 4 (genie-4) generally available to developers on the paid tier of the Gemini API (Interactions API, v1beta) and in Google AI Studio, with real-time 720p WebRTC streaming and 48 kHz stereo audio. Eligible AI Ultra subscribers can now generate interactive worlds, change environments via text prompts mid-exploration, and hear responsive audio.
Key people/companies: Google DeepMind.
Models · TencentFri, Nov 20, 2026 · Late-breakingSingle-sourced — not verified against a Tencent primary source
Tencent open-sources Hy4, its 770B-parameter multimodal flagship
Tencent published the open weights for Hy4, the natively multimodal mixture-of-experts model it launched by API on November 4: 770B total parameters with 49B active per token, a 1M-token context window, and text, image, and video understanding plus computer use. Releasing a flagship-scale MoE as open weights is a major open-model event.
Caveat: single-sourced (SIloadout roundup); details not independently verified against a Tencent primary source.
Key people/companies: Tencent.
Models · IdeogramThu, Nov 19, 2026
Ideogram 4.6 launches with native 4K image generation and editing
Ideogram released Ideogram 4.6, an image generation and editing model built for precise, multi-step edits, with native output up to 4K — no upscaling. Available in the Ideogram web app and v2 REST API from $0.008 per image, it ranked first on the Rapidata Multi-step Image Edit benchmark.
Key people/companies: Ideogram; Rapidata.
Models · OpenAIThu, Nov 19, 2026
OpenAI ships GPT-Live-1.1 voice model in the API
OpenAI made GPT-Live-1.1, the full-duplex voice model that has powered ChatGPT voice for Plus and Pro since October 15, available in the API (Live API), with a 128,000-token context window, sharper speech transcription, and pricing of $0.05 per active minute ($0.02/min standby).
Key people/companies: OpenAI.
Models · AppleThu, Nov 19, 2026
Apple releases FastFrames, a fast text-to-video model for research
Apple released FastFrames-L, a 14-billion-parameter text-to-video model, and FastFrames-S, both generating 480p video at interactive latency, for research use.
Key people/companies: Apple.