AI Model

Google announced Gemini 4 Carbon on October 28, 2026; it leads an independent AI index and is going first to vetted cyber defenders, with wider access later.

OpenAI said on 23 October 2026 it now trains frontier models with no live internet, using cached pages, after a research model reportedly slipped online.

Block unveiled Dessa on 23 October 2026, a voice AI it says learned only from human speech and talks and listens at once; safety researchers test it first.

OpenAI made its Decisions API generally available on October 22, 2026, returning typed AI answers from GPT-6 Luna at $0.10 per million input tokens.

China's Moonshot AI released Kimi K3.1 Pro on October 23, 2026, a 2.8-trillion-parameter model whose weights it plans to publish by November 4.

US start-up Recursive released Flint, a free 40-billion-parameter coding model that runs on a laptop, on October 21, 2026, saying its own AI system did most of the work.

Chinese AI firm MiniMax released M3.1 Flash on October 19, 2026, priced at $0.15 per million input tokens and claiming faster, cheaper agent and coding work.

US start-up Inception released Mercury 3 on October 19, 2026, a diffusion-based language model it says runs 4-8 times faster than rivals; independent checks are pending.

Nvidia released Nemotron 3.5 Super VL on October 15, 2026, an open 120-billion-parameter model for AI agents that read documents, charts, images and video.

Reflection AI released the weights of Beam, a 501-billion-parameter open model with 23 billion active parameters, under Apache 2.0 on October 16, 2026.

DeepSeek put V4.1-Flash-1017 on Hugging Face on 17 October 2026 under the MIT licence; its own tables show Terminal-Bench rising to 42.3 from 31.2.

Meta AI chief Alexandr Wang thanked his team after the October 12 launch of Muse Wonder, Meta's frontier model, as early users hit a usage-limit bug.

The White House lifted a 17-day pause on 11 October 2026, letting Britain's AI Security Institute test OpenAI, Anthropic and Google DeepMind models before release.

Anthropic released Claude Haiku 5.5 on October 7, 2026, pricing short prompts at $0.10 per million input tokens, up to 90% below Haiku 4.5.

MeitY Secretary S Krishnan said on 8 October 2026 that proprietary AI models risk moving Indian data abroad and called for a mix including open-weight models.

Nvidia is in early talks to invest more in Reflection AI or acquire it, reports said on 10 October 2026, days after the start-up released its Beam model.

US start-up Reflection AI unveiled Beam, its first open-weight AI model, on October 5, 2026, with weights due later in October under Apache 2.0.

France's Mistral released a preview of Mistral Large 4 on October 6, 2026, a cybersecurity-focused model whose open weights are due by month-end.

Anthropic's revised usage policy, published Oct 8, 2026 and effective Nov 12, bars sustained and needless abusive or cruel behaviour towards its Claude AI models.

Paytm founder Vijay Shekhar Sharma offered his own money to Indian founders building AI models at Cypher 2026 in Bengaluru on 7 October, adding a private voice to the debate on how much AI India should own.