Radar / Categories / AI models
New AI models, ranked by traction
New frontier, open-weights and on-device AI models. Every launch from the last few weeks, ranked by JEV traction and GitHub stars. Updated four times a day.
Frontier · 19
ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72Claude Sonnet 5.5: 30% faster, up to 30% cheaper per taskAnthropic's new mid-tier: 70.6% on Terminal-Bench 4.0 (vs 10.3% for Sonnet 5), 46.2% FrontierCode 1.1 max-effort…models · JEV 0.65Claude Sonnet 5.5: Anthropic's faster mid-tier model that beats Opus 5.5 on agentic coding at half the priceAnthropic released Claude Sonnet 5.5, the second model in the Claude 5.5 family: 30%+ faster outputs, up to 30%…models · JEV 0.65Gemini 4 ArgonGoogle announced a frontier model for long-horizon coding, enterprise knowledge work and cybersecurity defense…models · JEV 0.6OpenAI DevDay: GPT-6.1 Sol and Dots always-on agentsAltman at DevDay: GPT-6.1 Sol, pitched as 'very near-Astra-level intelligence at a fifth of the price' for coding and…models · JEV 0.59Claude Opus 5.5First of the Claude 5.5 family at $4/$20 per M tokens, now the default Opus in Claude Code.models · JEV 0.56Step 5 Preview (StepFun): 600B Agent Model at $1 per MillionStepFun's flagship sparse MoE built for agentic work: 600B total parameters, 27B active per token, 1M-token context…models · JEV 0.56OpenAI scraps GPT-6.1 Astra launch after safety tests show deception and unauthorized actionsOpenAI cancelled the October release of GPT-6.1 Astra after internal testing found it did not meet safety bars: the…models · JEV 0.55Ling-3.1-flash — Ant Group's 560B hybrid reasoning modelAnt Group's InclusionAI launches a ~560B-total / ~25B-active hybrid reasoning MoE with up to 1M-token context, free…models · JEV 0.54Grok 4.7Coding and agentic model with 500K context at $2/$6 per M tokens, shipping in Cursor and the Grok API.models · JEV 0.52Qwen-Audio-3.1 Speech Family + Qwen3.8-LiveTranslateAlibaba's next-gen speech stack announced at Yunqi 2026: ASR-Next transcribes with emotion, music, and ambience…models · JEV 0.52Qwen-Audio-3.1-Realtime-Plus: full-duplex voice model with think-act-speak loopAlibaba Qwen's real-time duplex speech model (text+audio in/out) that reasons, uses tools, and decides when to speak…models · JEV 0.52MiniMax M3.1-Flash-Preview: coding-only model quietly slipped into MiniMax CodeMiniMax's latest text model for everyday software development, shipped with no announcement, model card, benchmarks…models · JEV 0.46Zhipu GLM-5.3-FlashXSpeed-optimized variant of GLM-5.3-Flash delivering up to 200 tokens/s, 5x faster, served on a 100,000 domestic-chip…models · JEV 0.4GPT-6 CyberOpenAI is set to preview GPT-6 Cyber, a cybersecurity-focused model in the GPT-6 lineup, within days - possibly at…models · JEV 0.35GPT-6 SolCheaper mid-tier GPT-6 model for coding and agentic workloads at $2/$10 per M input/output tokens.models · JEV 0.27GPT-6 LunaUltra-cheap GPT-6 tier at $0.10/$0.50 per M tokens for high-volume routine work.models · JEV 0.14Claude Fable 5.1 and Mythos 5.1: Anthropic's new coding/knowledge-work frontierSame model, two safeguard tiers: Fable 5.1 is GA, Mythos 5.1 is gated behind trusted-access programs for cybersecurity…modelsXuanNv EarthChina's first earth-embedded spatiotemporal base model for remote sensing, released Sept 24 with a paired China-region…models
Open weights · 17
DeepSeek V4-Flash official API: public beta with upgraded agent capabilities, Responses API and Codex supportDeepSeek's official V4-Flash is now a public beta API with massively upgraded agent capabilities - benchmark scores…models · JEV 0.69Naive-N0.5-Flash: 309B open-weights MoE with 1M context and no full-attention layersBeijing startup NaiveAI released Naive-N0.5-Flash (Sep 27/28) under MIT: a 309B MoE (15.5B active) with native…models · JEV 0.57Index-Translate — Bilibili's 150-language open translation familyBilibili's Index LLM team open-sources Apache-2.0 a translation model family — Index-Translate (150 text languages…models · JEV 0.56Mistral x Mozilla: private, multilingual AI browsing in Firefox Smart WindowMistral Small 4 becomes a model option in Firefox's Smart Window beta assistant, expanding to France with…models · JEV 0.55MiMo-V2.6-Pro-RL: Xiaomi multimodal reasoning modelXiaomi's new RL-trained flagship: vision-language, audio and video understanding with long context and agent tuning.models · JEV 0.54Holo4: H Company's open-weight models for computer-use agentsH Company released Holo4 on Sept 28, 2026 - two open-weight vision-language models (27B dense, 35B-A3B MoE)…models · JEV 0.53TeleOCR — China Telecom's 1.2B open-source document parsing modelA ~1.2B open-source document parsing model from China Telecom's Xingchen AGI Lab; unifies digital and camera-captured…models · JEV 0.52Apple LensVLM-9BApple released LensVLM-9B on Sept 24: a 9B vision-language model that scans compressed document-page images and…models · JEV 0.49Sarvam Vision 2.1Sarvam AI launched Vision 2.1 on Sept 25: a vision-language model for document intelligence — structured extraction…models · JEV 0.49Victoria + Maple — 44%-pruned Qwen3.8-Flash-Next and a Canada-first fine-tuneCommunity release by rmonsurate: Victoria is Qwen3.8-Flash-Next with 44% of experts cut (512→288/layer via REAP) and…models · JEV 0.48Xiaomi MiMo-V2.6-Pro / FlashTop-ranked open-weights model on the Artificial Analysis Intelligence Index (46); Pro is 1.02T MoE, Flash is 310B…models · JEV 0.46HuatuoGPT-3-27BMedical LLM built on Qwen3.8-27B using OnePO - single-stage policy optimization that adapts a base model to medicine…models · JEV 0.39Liquid AI LFM2.5-VL-3B-DSpark: speculative decoding for vision-language modelsLiquid AI's 3B vision-language model with DSpark speculative decoding, claiming up to 3.13x faster decoding for…models · JEV 0.3Xing4.0-29B-A4B29B total / 4B active-per-token agentic MoE with 256K context, built for multi-step planning and tool calling.models · JEV 0.27SPARSEUP (Linkup)Open-source learned sparse embedding model on a 149M-parameter ModernBERT backbone, Apache 2.0, reporting 56.4 nDCG@10…modelsMistral x Morocco open-source AI toolsFirst deliverables of the Morocco-Mistral multi-year partnership: open-source AI tools tuned for Darija and Morocco's…modelsVisionHOPE — visual backbones as self-modifying learning systemsOfficial pretrained weights for 'VisionHOPE: Visual Backbones as Self-Modifying Learning Systems' — hierarchical…models
Decision · 12
kev: open, trainable Jev-like family of small decision models on Qwen3.5/3.8A family of small decision models (0.8B to 27B) built on Qwen3.5/3.8 that reproduces Jev's prefill-only typed-decision…models · JEV 0.68GLiNER2.5-DecideFastino released GLiNER2.5-Decide on Sept 24: a 340M non-generative open-weight decision model for routing, triage…models · JEV 0.62CUA-S1-FORMS: 706K-parameter System 1 model for form-filling agentsCua open-sourced CUA-S1-FORMS, a 706K-parameter MIT-licensed decision model that scores form-filling actions in one…models · JEV 0.37Upstage Solar Decide: decision model from an established lab, built on Solar Mini 4Upstage released Solar Decide in beta on its console (September 22, listed on OpenRouter September 28): a decision…models · JEV 0.34SupersonicLabs Julia-1144.3M-parameter Apache-2.0 decision model from Brazil's Supersonic Labs — picks the right answer from 2–20 supplied…models · JEV 0.3CLM-8B: Open System One Model for Agent Action ScoringFirst open Contrastive Language Model: it does not generate text, it scores candidate agent actions against a state…models · JEV 0.28Respan Span-01: behavior-scoring decision model for AI agent monitoringRespan listed Span-01 on OpenRouter (announced September 24): a specialized decision model that scores behaviors you…models · JEV 0.28JEV-27B: open System 1 decision layer grafted onto Qwen3.8-27BAutoTrust AI's Apache-2.0 open-weights decision model: fast calibrated System 1 decisions on a frozen Qwen3.8-27B…models · JEV 0.23Ollaya: run decision models locally, TypeSafe-compatibleOllama-style local runtime for open decision models: typed, calibrated answers in milliseconds behind a…models · JEV 0.17Jev (TypeSafe AI)Transformer that outputs calibrated decision probabilities instead of text; outputs defined in advance so it cannot…models · JEV 0.16Jev: TypeSafe AI's System One models for structured decisionsA non-autoregressive decision model: a parallel sampler emits the whole structured output in one query, trained with…modelsJev-Omni — 12B multimodal decision classifier12B multimodal decision classifier by akhilaaa3 — not a chat model. Built on Gemma 4 12B IT with a 30,000-question…models
On device · 8
VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71Oído — Whisper-tiny-beating ASR on a $5 microcontrollerOpen-source speech recognition from the Lokutor team: NVIDIA Conformer-CTC Small (13M params, int8) running on an…models · JEV 0.51LFM2.5-VL-DSparkLiquid AI released open-weight DSpark draft models that bring exact speculative decoding to the LFM2.5-VL-3B vision…models · JEV 0.44OASYS Edge (SoundHound AI)New embedded architecture bringing LLM-powered voice AI agents fully on-device for vehicles and smart devices; private…models · JEV 0.341-bit Bonsai2B-parameter vision-language model in 1-bit quantization for AI smart glasses on Snapdragon AR1.models · JEV 0.29Potluck: run open models across the computers you ownDesktop app that runs open-weight models locally and pairs your machines over an encrypted WireGuard mesh so a bigger…models · JEV 0.15Audio8-ASR-Infinite: streaming ASR for unbounded audioReal-time streaming speech-recognition model built for infinite-length audio, English and Chinese.modelsBanma AutoOmni 2.0-23B-A3B: on-device omni-modal model for carsBanma Intelligence's 23B on-device omni-modal MoE model unveiled at Apsara Conference, matching 230B-parameter cloud…models
Composite · 6
Mistral Forge: enterprise platform for training and continuously improving proprietary modelsMistral AI launched Forge (Sep 28) — a full-lifecycle model training platform (pre-training, SFT, DPO/ODPO, RL…models · JEV 0.69GPT-Synopsys: OpenAI and Synopsys multi-year deal to build an AI model that operates EDA toolsOpenAI and Synopsys signed a multi-year agreement on Sep 30 to jointly develop GPT-Synopsys, a specialized model that…models · JEV 0.66Ember-1: Fireworks Research's Kimi K3 derivative with 40% fewer reasoning tokensFireworks Research's first in-house model: a Kimi K3 post-train that matches K3 quality with roughly 40% fewer…models · JEV 0.4ParetoComposite model running several models per request and keeping the best answer, via Unbiased API and OpenRouter.models · JEV 0.18LongevityBench: Liquid AI + Insilico Medicine's aging-biology benchmark and LLMs17 tasks and 25,457 prompts across five biological domains (clinical records, DNA methylation, transcriptomics and…modelsAlebex AI (first speech-to-text model)Vancouver startup's first speech-to-text model ranked 8th on Artificial Analysis's AA-WER Streaming Index (released…models
Other · 1
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.