# Models

[https://timesof.ai/category/models](https://timesof.ai/category/models)

Model releases, benchmarks, capability jumps, and lab research milestones.

## Coverage

- [Google's Cheaper Gemini 4 Hydrogen Scores 51 on an Independent Index, Though at Its Top Setting It Still Ranks 16th](https://timesof.ai/2026/11/google-gemini-4-hydrogen-everyday-agent-model-1-50-7-50) — Google's Gemini 4 Hydrogen targets coding and agent work at $1.50/$7.50 per million tokens. Artificial Analysis scores it 51 (high) and 54 (xhigh).
- [MiniMax's Open-Weight H3-Image Makes 4K Pictures in Seconds, Under a Licence That Shuts Out the US, EU, UK and Korea](https://timesof.ai/2026/11/minimax-h3-image-open-weights-4k-license-excludes-us-eu-uk-korea) — MiniMax's open-weight H3-Image (33.1B DiT, six steps) claims 4K in under 10s. Licence excludes the US, EU, UK and Korea; $1M revenue cap.
- [Runway Opens the Weights of Its Video-Trained Robot Model, the Top World Action Model on Nvidia's Leaderboard but Fifth Overall](https://timesof.ai/2026/11/runway-praxis-1-open-weights-robot-model-robolab) — Runway's 27B Praxis-1 world action model is open-weight. It leads world action models on RoboLab-120 at 46.2% but ranks fifth overall on the leaderboard.
- [Google's New Gemini Model Doesn't Write Answers, It Picks Them: Decision 1 Scores Up to 1,000 Questions for $0.10 per Million Tokens](https://timesof.ai/2026/11/google-gemini-decisions-api-decision-1-closed-set-choices) — Google's Gemini Decisions API scores yes/no, choice and rubric questions with confidence scores at $0.10/M input tokens. An independent index ranks it 9th of 18.
- [Alibaba Previews Qwen4-Max for Two Weeks, Subscription-Only and With No Independent Scores Yet](https://timesof.ai/2026/11/alibaba-qwen4-max-preview-1m-context-token-plan-only) — Qwen4-Max preview opens Nov 11 with image/video input and 1M context, via Alibaba's Token Plan only; it retires Nov 25. No independent benchmark listing yet.
- [Meta's Muse Ember Lists at Haiku's and Luna's Price; Its Edge Is Spending Fewer Tokens per Task](https://timesof.ai/2026/11/meta-muse-ember-cheapest-agent-model-haiku-luna-per-task) — Meta's Muse Ember lists at $0.10/$0.50, like Haiku 5.5 and GPT-6 Luna. Artificial Analysis scored it 36, level with Haiku, at about a third of the cost per task.
- [Anthropic Cuts Prices With Claude Fable 5.2 and Mythos 5.2, Says It Held Back the Model's AI-Research Skills](https://timesof.ai/2026/11/anthropic-claude-fable-mythos-5-2-price-cut-cisco-ibm) — Anthropic's Fable 5.2 and Mythos 5.2 cost $8/$40 per million tokens. It says it capped AI-research skills at Opus 5.5 level. Benchmarks are Anthropic's own.
- [Baidu Ships ERNIE 5.5 Flash Preview, a Post-Trained 280B MoE on Its Raw Base, API-First With Apache 2.0 Weights Promised](https://timesof.ai/2026/11/baidu-ernie-5-5-flash-base-280b-moe) — Baidu released ERNIE 5.5 Flash Preview (280B/14B MoE, 1M context) on Qianfan, with Apache 2.0 weights promised soon. Vendor claims only so far.
- [Thinking Machines Releases Inkling 2 Small's Weights Under Apache 2.0, a Week After Its Closed Launch on Tinker](https://timesof.ai/2026/11/thinking-machines-inkling-2-small-tinker) — Thinking Machines published Inkling 2 Small (594B/17B MoE) weights on Hugging Face under Apache 2.0. The NVFP4 checkpoint needs about 500GB of GPU memory.
- [Google Makes Gemini Omni 1.2 Flash Generally Available With 15-Second Clips and a 30% Price Cut](https://timesof.ai/2026/11/google-gemini-omni-1-2-flash-generally-available) — Gemini Omni 1.2 Flash is GA on API and Google Cloud: 15-second clips, $0.07/s at 720p (−30%), multi-turn edits, SynthID, Provisioned Throughput.
- [Ricursive Releases Ripple-65B Preview, an Apache 2.0 Model It Says Tops US Open-Weight Coding Scores for $32 Million](https://timesof.ai/2026/11/ricursive-ripple-65b-preview-apache-open-weights-coding) — Ricursive's Apache 2.0 Ripple-65B Preview reports 80.52 on TerminalBench 2.1, above larger US open models, at a stated $32M full cost. Self-reported.
- [Tencent's Hy4 Adds Vision and Computer Use, and Tencent Says the Open Weights Follow in Days](https://timesof.ai/2026/11/tencent-hy4-vision-computer-use-open-weights-coming) — Tencent's Hy4 adds vision and computer use, trained on Blender and CAD, at preview prices. Weights and system cards are promised in days; scores are Tencent's own.
- [Meituan's LongCat-2.5 Is a 1.6 Trillion-Parameter Agent Model Trained on Chinese Chips; Its Own Charts Put It Behind Claude Opus 5 on the Hardest Coding Test](https://timesof.ai/2026/11/meituan-longcat-2-5-1-6t-multimodal-domestic-accelerators) — Meituan's LongCat-2.5 has 1.6T parameters, 48B active, trained on Chinese accelerators. Vendor benchmarks trail Claude Opus 5 on Terminal Bench v4.0.
- [NVIDIA's Open Nemotron 3.5 Transcribe Puts Transcripts, Speaker Labels and Translation in One Model, for Recorded Audio Only](https://timesof.ai/2026/10/nvidia-nemotron-3-5-transcribe-open-asr-granary-v2) — NVIDIA's open Nemotron 3.5 Transcribe handles transcription, translation and speaker labels in one 30B model for recorded audio. Leaderboard claims are NVIDIA's.
- [Gemini 4 Carbon Opens to Paying Developers at $2/$10 Until November 30, but Now Sits Fourth on Artificial Analysis Behind Claude Fable 5.2](https://timesof.ai/2026/10/google-gemini-4-carbon-fairwind-cyber-frontier) — Gemini 4 Carbon is now on the paid Gemini API and AI Ultra at $2/$10 until Nov 30, then $4/$20. Artificial Analysis ranks it fourth, behind Claude Fable 5.2.
- [Liquid AI Releases LFM3, Six Open-Weight Models Around 2B Parameters Covering Text, Vision, Audio and Omni Inputs](https://timesof.ai/2026/10/liquid-ai-lfm3-2b-open-weight-multimodal) — Liquid AI's LFM3 family launches with six 2B-class open-weight models for text, vision, audio and omni input. Its license caps commercial use at $10M annual revenue.
- [Google DeepMind's Genie 4 World Model Generates Sound and Dialogue With Its Worlds and Takes Text Edits in Real Time, Starting With Trusted Testers](https://timesof.ai/2026/10/google-deepmind-genie-4-world-model) — Genie 4 generates interactive worlds with native sound and dialogue, takes real-time text edits and uses Gemini 3.8 Flash to steer scenes. It starts as a research preview.
- [Ant Group's inclusionAI Releases Ling-3.1-flash-1019, a 560B-Parameter Open-Weight Model Under the MIT License](https://timesof.ai/2026/10/ant-ling-3-1-flash-1019-open-weights-mit) — Ant Group's inclusionAI releases Ling-3.1-flash-1019, a 560B-parameter mixture-of-experts model with 25B active parameters and a 1M-token context, under the MIT license.
- [OpenAI Prices GPT-6.1 Terra at $1 and $5 per Million Tokens, and Its System Card Reports 29.6% Restriction Workarounds in a Test Without Safeguards](https://timesof.ai/2026/10/openai-gpt-6-1-terra-launch-pricing-system-card) — OpenAI says GPT-6.1 Terra costs 50% less than GPT-5.6 Terra. Its system card reports a 29.6% workaround rate in one test run without safeguards.
- [Moonshot Publishes Kimi K3.1 Pro Open Weights, a 2.8-Trillion-Parameter Model With Cyber Training Data Removed](https://timesof.ai/2026/10/moonshot-kimi-k3-1-pro-weights-nov-4-cyber-data-removed) — Moonshot published Kimi K3.1 Pro weights on Hugging Face: 2.8T-parameter MoE, 104B active, 1M context, under a custom licence. Benchmarks self-reported.
- [Block's AI Lab Unveils Dessa, a Voice Model It Says Learned to Talk Only From Recorded Speech](https://timesof.ai/2026/10/block-dessa-full-duplex-speech-model-trained-only-on-speech) — Block says its Dessa voice model learned only from recorded conversation, with no text. No benchmarks or release date yet; safety researchers get early access.
- [Recursive Releases Flint, an Apache-2.0 Open-Weight Coding Model for Laptops That It Says Its Automated-Research System Helped Build](https://timesof.ai/2026/10/recursive-flint-40b-open-weight-laptop-coding-model-apache-2) — Recursive's Flint, a 40B/2.5B-active MoE coding model, is downloadable under Apache 2.0. Recursive claims sub-50B MoE benchmark leads and >125 tok/s on an M4 Pro.
- [Mistral Launches Voxtral 3 Speech Models and a Voice-Agent API From $0.04 a Minute](https://timesof.ai/2026/10/mistral-voxtral-3-speech-models-voice-agent-api-eu-hosted) — Mistral Voxtral 3: STT 50+ languages, TTS 35+ languages, voice-agent API from $0.04/min. EU/US/global or self-deploy. Benchmarks are Mistral's.
- [Update: MiniMax Publishes M3.1 Flash Open Weights Under a Community Licence That Limits Commercial Use; AA Scores It About 42](https://timesof.ai/2026/10/minimax-m3-1-flash-aa-42-weights-promised-oct-23) — MiniMax M3.1 Flash weights are on Hugging Face under a community licence that limits commercial use: open weights, not open source. AA scores it about 42.