New AI Models — August 2026
Every model release from our AI news in August 2026, newest first.
Model changes only: new models, open weights, API access, first platform listings and leaderboard ranks. No product features, pricing or single benchmark scores.
65 releasesLast updated: August 31, 2026
- StartLux 27BStartLuxLeaderboardStartLux's 27B local model takes second on CAICT's MCP test, beating DeepSeek-V4-Flash
- GLM-5.3Zhipu AIOpen weightsZ.ai opens GLM-5.3 weights for coding and vulnerability huntingGLM timeline →
- Hunyuan Hy4 previewTencentOpen weightsTencent open-sources Hunyuan Hy4 preview: 770B total, 49B active
- GLM-5.3-FlashZhipu AILeaderboardGLM-5.3-Flash tops OpenRouter usage, served on domestic chipsGLM timeline →
- Gemini Omni 1.1 FlashGoogleNew modelGemini Omni 1.1 Flash Adds Scene and 4K ControlsGemini timeline →
- WALL-SSX Square RobotNew modelWALL-SS Simulates 60-Second Robot Tasks
- Thomson-1Thomson ReutersNew modelThomson Reuters Reportedly Launches Qwen-Based Thomson-1
- GEN-1.5Generalist AINew modelGEN-1.5 learns tasks from seconds-long demos
- SHADOW 250M InstructUndisclosedOpen weightsSHADOW 250M Fits Into a 60MB Deployment
- TenetHarveyNew modelHarvey Tenet Post-Trains Kimi K3 for Legal Agents
- Ox AlphaUndisclosedNew modelAnonymous Ox Alpha Offers a One-Million-Token Window
- Wan3.0AlibabaNew modelWan3.0 Generates 30-Second Video from Documents
- LFM2.5-DSparkLiquid AINew modelLFM2.5-DSpark lifts inference throughput up to 3.18x via speculative decoding
- Qwen-UI-AgentAlibabaNew modelAlibaba releases Qwen-UI-Agent, a foundation model for GUI agentsQwen timeline →
- HappyShrimpAlibabaNew modelAlibaba launches HappyShrimp AI music model
- Gemini 3.7 FlashGoogleNew modelGoogle ships Gemini 3.7 Flash at half price, three weeks after 3.6Gemini timeline →
- Palmyra X6WriterNew modelWriter launches Palmyra X6, post-trained on Z.ai's GLM-5.2
- OlmoEarthAi2Open weightsAllenAI opens OlmoEarth embedding exports, with code and weights released
- Qwen3.8-2.4T-A95BAlibabaOpen weightsAlibaba open-weights Qwen3.8-2.4T-A95B, its first Max-tier weightsQwen timeline →
- DeepSeek-V4-Pro-0813DeepSeekNew modelDeepSeek V4 Pro goes live with a 1M-token tierDeepSeek timeline →
- SL2TGoogleNew modelGoogle DeepMind ships sign-language-to-text model SL2T
- LFM2.5-VL-3BLiquid AINew modelLiquid AI releases edge vision model LFM2.5-VL-3B
- MAI Code 1.1 FlashMicrosoftNew modelMicrosoft's MAI Code 1.1 Flash trails DeepSeek on price and performance
- Ishigaki-IDSONESTRUCTIONNew modelONESTRUCTION builds construction-BIM foundation model Ishigaki-IDS with AWS
- Nemotron 3.5 LightningNVIDIAOpen weightsNvidia open-sources Nemotron 3.5 Lightning, a fast 30B MoE
- Daybreak RedOpenAIPlatformOpenAI's Daybreak cyber models arrive on Amazon Bedrock
- Daybreak BlueOpenAIPlatformOpenAI's Daybreak cyber models arrive on Amazon Bedrock
- LTX-2.5LightricksOpen weightsLTX-2.5 ships as an open-weights video world model that runs on one desk
- Muse GlimmerMetaOpen weightsMeta open-sources Muse Glimmer, a 30B model for one GPU
- Needle 2CactusNew modelCactus ships Needle 2, a 14MB on-device agentic model
- VoiceChat 11BNVIDIAOpen weightsNVIDIA open-sources full-duplex speech model VoiceChat 11B
- Isaac 28BPokee AINew modelPokee AI ships Isaac 28B with a 10M-token context on a single GPU
- Wan 3.0AlibabaNew modelAlibaba's Qwen app adds five free features; Wan 3.0 enters public test
- Cosmos 3NVIDIAOpen weightsNVIDIA's open Cosmos 3 world models span 64B/16B/4B
- WeatherNextGoogleOpen weightsDeepMind open-sources WeatherNext for cyclone forecasts
- Xiaomi-Robotics-1XiaomiOpen weightsXiaomi open-sources Xiaomi-Robotics-1, a 5B robot model
- Muse Spark 1.2MetaNew modelMeta ships Muse Spark 1.2 and the Muse Code terminal coding agent
- Doubao 2.0ByteDancePlatformSAIC Roewe Jiayue 07 to run Doubao 2.0 and Momenta's R7 world model
- Alpamayo 2 SuperNVIDIAOpen weightsNVIDIA open-sources Alpamayo 2 Super: a 34B vision-language-action model for autonomous driving
- MAGI-2-previewSand.aiOpen weightsSand.ai open-sources MAGI-2-preview: a 114B/6B-active MoE video model
- ShieldstralMistral AIOpen weightsMistral releases Shieldstral, a 3B open-weight multimodal safety classifier
- LFM2.5-2.6BLiquid AINew modelLiquid AI ships LFM2.5-2.6B for on-device agents
- DeepSeek V4-FlashDeepSeekLeaderboardDeepSeek V4-Flash tops OpenRouter's weekly token rankingDeepSeek timeline →
- MiniMax H3MiniMaxOpen weightsMiniMax open-sources H3, first open model to top a video rankingMiniMax timeline →
- SenseNova U1.5-LiteSenseTimeOpen weightsSenseTime open-sources a preview of SenseNova U1.5-Lite (Chinese-language source)
- Instella-MoEAMDOpen weightsAMD ships the fully open Instella-MoE: 16B total, 2.8B active
- DoubaoByteDancePlatformTesla China adds ByteDance's Doubao — its first third-party in-car LLM
- Seedance 2.5ByteDancePlatformSeedance 2.5 video model arrives in the creation tool "Xiaoyunque"