Google announces Gemini 4 Carbon
Google's frontier model announced on 28 October 2026, released first to cyber defenders and reporting leading scores on intelligence, coding and cybersecurity benchmarks.
Total 75 · Clear filters
Google launches Gemini 4 Argon
Google gave Gemini 4 Argon to trusted security partners in September 2026 and released it worldwide on 13 October, opening its newest frontier model to developers, enterprises and subscribers while naming Gemini 4 Carbon as the next model in the family.
IBM releases Granite 4.2 with built-in reasoning modes
IBM released the Granite 4.2 family on 25 August 2026, with native reasoning that streams inside think tags and can be switched between full, low-effort and non-thinking modes per query.
Alibaba releases Qwen3.8-Max, a 2.4-trillion-parameter model
Alibaba released Qwen3.8-Max, a 2.4-trillion-parameter model, as a cloud service on 3 August 2026, and published open weights for a variant of it on 12 August 2026.
NVIDIA releases Nemotron 3 Ultra, a 550B open model
On 4 June 2026 NVIDIA released Nemotron 3 Ultra, the largest model in its Nemotron 3 family, with 550 billion total parameters and 55 billion active per token.
Microsoft AI launches seven MAI models led by MAI-Thinking-1
On 2 June 2026 Mustafa Suleyman announced seven new in-house MAI models at Microsoft Build, led by the reasoning model MAI-Thinking-1. The group also included coding, image, transcription and voice models, and a healthcare collaboration with Mayo Clinic.
DeepSeek releases the DeepSeek-V4 preview with a 1M-token context
On 24 April 2026 DeepSeek released the DeepSeek-V4 preview, two open mixture-of-experts models with a 1M-token context that became the default across its official services. Both models support thinking and non-thinking modes.
OpenAI releases GPT-5.5
OpenAI released GPT-5.5 Thinking and GPT-5.5 Pro to paid ChatGPT users on 23 April 2026, then extended the models to Codex and the API, reporting strong benchmark results and later making a faster GPT-5.5 Instant the default.
Meta Superintelligence Labs releases Muse Spark
On 8 April 2026 Meta Superintelligence Labs released Muse Spark, the first model from the lab and the first of the Muse series. It is closed: its weights were not released, and it was offered through a private API preview.
Mistral AI releases Mistral Small 4
On 16 March 2026, Mistral AI released Mistral Small 4, a 119B mixture-of-experts model under Apache 2.0 that combines reasoning, image input and agentic coding in one model.
Microsoft releases Phi-4-reasoning-vision-15B
On 4 March 2026 Microsoft released Phi-4-reasoning-vision-15B, an MIT-licensed open-weight model that combines a reasoning backbone with a vision encoder and chooses how to answer each task.
Google releases Gemini 3.1 Pro
On 19 February 2026 Google released Gemini 3.1 Pro in preview, a model that scored 77.1% on the ARC-AGI-2 reasoning benchmark — more than double Gemini 3 Pro — and was made available the same day in the Gemini app, the Gemini API and Vertex AI.
xAI releases Grok 4.20 with a multi-agent mode
Grok 4.20 was released in February 2026, the month xAI became a SpaceX subsidiary. Its multi-agent version, in which agents collaborate on deep research, went live in the API in March 2026.
Baichuan releases the open Baichuan-M3 medical model
On 13 January 2026 Baichuan released Baichuan-M3, a new-generation open-source medical-enhanced LLM, at a media briefing.
OpenAI releases GPT-5.2 after a "code red"
On 11 December 2025 OpenAI released GPT-5.2 in Instant, Thinking and Pro versions, billing it as stronger at professional work and reporting scores that, on OpenAI's GDPval evaluation, matched or beat human experts.
Tencent releases HY 2.0 with Think and Instruct versions
On 5 December 2025 Tencent released HY 2.0 in Think and Instruct versions, a 406-billion-parameter mixture-of-experts model with a 256K context, integrated into Yuanbao and ima.
Amazon launches Nova 2 and Nova Forge at re:Invent 2025
On 2 December 2025, at AWS re:Invent, Amazon released the Nova 2 family, led by the reasoning model Nova 2 Lite. The same event introduced Nova Forge, made Nova Act generally available and launched Trainium3 UltraServers.
DeepSeek releases V3.2 and gold-medal-level V3.2-Speciale
On 1 December 2025 DeepSeek released V3.2, the official successor to V3.2-Exp, and V3.2-Speciale, a version built to maximize reasoning. Both had their weights published on Hugging Face, and V3.2-Speciale was reported to reach gold-level results at several international competitions.
DeepSeekMath-V2 reaches gold-medal level on IMO 2025 problems
On 27 November 2025 DeepSeek released DeepSeekMath-V2, a model trained to check its own proofs. DeepSeek reported gold-level scores on IMO 2025 problems and 118 of 120 points on Putnam 2024, the latter with scaled test-time compute.
Ai2 releases Olmo 3 with open reasoning models
On 20 November 2025 Ai2 released Olmo 3 in 7B and 32B sizes, in Base, Think, Instruct and RL Zero variants. Ai2 called Olmo 3-Think 32B the strongest fully open thinking model.
Google launches Gemini 3
On 18 November 2025 Google announced Gemini 3, starting with Gemini 3 Pro in preview alongside a harder-reasoning Deep Think variant, and reported top scores on public benchmarks and same-day availability in Search.