# [AINews] Here are 6 Clones of Jev in 2 days > Source: > Published: 2026-09-19 05:48:28+00:00 We covered [Jev’s launch on Wednesday](https://www.latent.space/p/ainews-jev-a-system-one-model-that), and they have completely taken over the timeline, with 36M views of their launch video (by comparison, [OpenAI’s Navier Stokes](https://www.latent.space/p/ainews-openai-reports-navier-stokes) result got [74M](https://x.com/OpenAI/status/2097374640582668336?s=20) views, and [Anthropic’s Fable 5](https://www.latent.space/p/ainews-anthropic-claude-fable-5-mythos?utm_source=publication-search) got [57M](https://x.com/claudeai/status/2064394146916229443?s=20) views) in just two days. It wasn’t open source[1](#footnote-1), so it invited tons of speculation and great demos and [examples](https://archerhume.com/posts/jevs-architecture-unmasked/?v=3) and [salty schmidhubers](https://x.com/george_onx/status/2100293114808119379?s=12) and [bad takes](https://x.com/theo/status/2100762304862384257), which of course only fed the hype. Here’s a list. The best guesses are ModernBert and Diffusion: - [Laya](https://github.com/NandhaKishorM/laya) : 421M params,[ModernBERT](https://ai.engineer/talks/YZHPEkfy2kc-1-ai-guardrails-unreasonable-effectiveness) -large encoder with two added transformer layers that score user-supplied options, PPO over sequence embeddings to output turn-by-turn conversion trajectories (probabilities from 0.0 to 1.0). - salty that he did not get recognition; claims RLCD without justification - confidence is entropy-based, not calibrated - [DiffusionGemmaJev:](https://github.com/vllm-project/vllm/pull/57250) tackling this from a Diffusion model basis.[Pretty close on benchmarks](https://x.com/mmastrac/status/2100626193943052784) - [Bespoke Nimble](https://x.com/madiator/status/2100990591215783946?s=20) : LoRA finetune of Qwen3.5-9B, using contrastive data curation. ([close but sllightly lower on benchmarks](https://x.com/madiator/status/2101128095831093579/photo/1) ) - [SemIf (fka OpenJev)](https://openjev.com/) ([HF](https://huggingface.co/AlexWortega/openjev) ): 4B and 35B causal Qwen3.5 backbone with a tiny three-class NLI classifier on the last token.[comparison vs Laya](https://www.reddit.com/r/LocalLLaMA/comments/1wjieap/comment/pajt99a/) - [Jevlike](https://github.com/vinnylarouge/jevlike) : 40K byte embedding lightweight option-attention model. Each candidate becomes a query that reads from a shared context representation, then receives a score. - [Kev-0.5B](https://x.com/jaredpalmer/status/2101028325472841920) : LoRA adapter + a small readout head on top of Qwen2.5-0.5B. Of course, not enough people are talking about [the data side, which is acknowledged to be 100% synthetic](https://x.com/completeskeptic/status/2100617775823966680?s=12). AI News for 9/17/2026-9/18/2026. We checked 12 subreddits, [544 Twitters](https://twitter.com/i/lists/1585430245762441216) and no further Discords. [AINews’ website](https://news.smol.ai/) lets you search all past issues. As a reminder, [AINews is now a section of Latent Space](https://www.latent.space/p/2026). You can [opt in/out](https://support.substack.com/hc/en-us/articles/8914938285204-How-do-I-subscribe-to-or-unsubscribe-from-a-section-on-Substack) of email frequencies! # **AI Twitter Recap** **Decision Models, Routing, and the “Jev” Wave** - **Discriminative models broke out as a new systems primitive** : The biggest technical conversation was around**Jev** , a non-generative decision model being positioned as a fast “**System 1** ” complement to LLMs.[@ankrgyl](https://x.com/ankrgyl/status/2100978416434786420) said it is now available as an eval model in Braintrust with**~400x lower scoring cost** versus prior setups, while[@gabepereyra](https://x.com/gabepereyra/status/2100990093691691382) highlighted calibrated-probability use cases like routing, citation selection, escalation, and legal ops decisions. The more architectural take came from[@hxiao](https://x.com/hxiao/status/2101001002816327867) , who argued Jev could pull tool calling, routing, and MCP-style decisions back from small generative LMs toward discriminative models;[@signulll](https://x.com/signulll/status/2101062047350096040) pushed the same idea further, framing this class as a near-zero-marginal-cost,**on-device judgment layer** for notifications, UI adaptation, and sensor-driven decisions. - **Open reproductions and ecosystem clones appeared immediately** :[@madiator](https://x.com/madiator/status/2100990591215783946) released**Bespoke Nimble** , an “open Jev” recipe built from a**LoRA fine-tune of Qwen3.5-9B** using**synthetic contrastive data curation** and constrained decoding. On its curated eval, the base Qwen improved from**66% to 90%** , versus**93% for Jev** , with a reported**100ms on H100** and local usability. At the smaller end,[@jaredpalmer](https://x.com/jaredpalmer/status/2101028325472841920) released**Kev-0.5B** , a tiny Jev-like model based on**Qwen2.5-0.5B** that can run on a MacBook Pro. The reaction split roughly along prior experience:[@MParakhin](https://x.com/MParakhin/status/2101036299721347073) noted post-ChatGPT users treated it like a revelation, while pre-GPT ML people were more puzzled by the hype. The substantive question raised by[@abacaj](https://x.com/abacaj/status/2101048462661845099) is the right one: a lot of demos emphasized**speed** more than**quality** , and there is still no standard benchmark for this category. - **The first compelling integrations were in browser/computer-use workflows** :[@levie](https://x.com/levie/status/2101007708044574906) demoed Jev classifying Box incident reports into escalation paths;[@ndrezn](https://x.com/ndrezn/status/2101046780989215005) showed browser use with LangChain + Jev and found it strong on tasks like the Wikipedia game and structured “folding laundry” workflows;[@cline](https://x.com/cline/status/2101056078872256935) shipped a plugin giving Jev a browser in Cline.[@hwchase17](https://x.com/hwchase17/status/2101054310037790814) explicitly called browser use the best Jev application he had seen so far. Net: this looks less like a chatbot story than a**workflow control-plane** story. **Agent Tooling, Coding Harnesses, and Claude Code Standards** - **AGENTS.md gained real momentum as a cross-tool convention** : The highest-signal product update here was[@trq212](https://x.com/trq212/status/2101009392611278961) announcing that**Claude Code v2.1.277** now checks for**AGENTS.md** when no**CLAUDE.md** is present, with config-level toggle support. That effectively acknowledges AGENTS.md as an emerging standard rather than a one-tool convention, and[@simonw](https://x.com/simonw/status/2101025043098812807) immediately noted the practical payoff: fewer shim files that just point one format to the other. - **Harness design is becoming a first-class variable in coding-agent performance and cost** :[@pidotdev](https://x.com/pidotdev/status/2100935860413673605) highlighted the**Harness Tax** analysis showing that a simple tool set—**read, write, edit, bash** —can reach the**Pareto frontier** on benchmark performance while reducing unnecessary spending. Relatedly,[@_akhaliq](https://x.com/_akhaliq/status/2101020560964866103) pointed to the paper*An Empirical Study of Harness Design for Coding Agents* , underscoring that benchmark outcomes are increasingly shaped by**harness structure** , context setup, turn budgets, and tool affordances rather than just the base model. This is consistent with[@dexhorthy](https://x.com/dexhorthy/status/2100900279021363244) ’s “software factory” argument that teams still need to**read the code** and deliberately design the human/agent interface. - **Model choice in software systems is bifurcating** : Several practitioners described a split between “frontier for planning, cheap for execution.”[@TheAhmadOsman](https://x.com/TheAhmadOsman/status/2101061704444682682) summarized one stack as**GPT 5.6 Sol XHigh** for planning,**GLM 5.3 Flash** for implementation, and**DeepSeek V4.1 Flash** for other tasks.[@kylebrussell](https://x.com/kylebrussell/status/2101028521044812109) reported an internal knowledge-base pipeline moving from**Opus → Sonnet → GLM 5.2 → GLM 5.3 Flash** , cutting spend by roughly**two orders of magnitude** since spring. Meanwhile[@theo](https://x.com/theo/status/2101062549722841452) argued that in real-world coding the payoff from stronger models like**Fable** and**Astra** is not just code quality, but a subtler productivity gain in execution and iteration. **Benchmarks, Recursive Self-Improvement, and Math Capability** - **RSI discussion got more precise about what is actually “recursive”** :[@TheTuringPost](https://x.com/TheTuringPost/status/2100769692877303863) offered a useful taxonomy: AI improving code or training methods is not, by itself, fully recursive if the surrounding improvement loop remains fixed. The key threshold is when AI can modify not just model internals, but**search strategy, experience generation, research tooling, and the improvement process itself** . That framing links well with[@HuaxiuYaoML](https://x.com/HuaxiuYaoML/status/2100959310624825688) ’s**RSI-Exam** update, where**GPT-6-astra** remains #1 at**0.5126** , with**Fable 5.1** entering at #2 with**0.4813** , and no model yet reaching the frontier-calibrated reference. - **Math benchmarks continued to fall to frontier models, but interpretation remains nuanced** :[@EpochAIResearch](https://x.com/EpochAIResearch/status/2100986494873989227) reported that another**FrontierMath open problem** was solved in an interactive session with**GPT-6 Astra** . Separately,[@SAIRfoundation](https://x.com/SAIRfoundation/status/2100976455123620089) launched**Open Math Model** , pitching open models and tools for mathematics shaped by the research community. Against the “verifiability explains math strength” narrative,[@steve47285](https://x.com/steve47285/status/2100998663254225391) shared an argument that**pretraining data** , not merely verifiable reward structure, is the main reason LLMs are so good at math and coding. The meta-point from[@sarahcat21](https://x.com/sarahcat21/status/2101023982258712725) is worth keeping: we need not just better benchmarks, but better**benchmark maintenance and audit tooling** . - **Computer-use benchmarks are still far from saturation** :[@ValsAI](https://x.com/ValsAI/status/2101014465781318072) launched**CUA-Bench** , testing real-time keyboard/mouse use across**6 games** (with**3 kept private** ) as a proxy for difficult human-easy tasks. Follow-up numbers from[@ValsAI](https://x.com/ValsAI/status/2101014471586243036) suggest this remains genuinely hard:**all frontier models score below 20%** . In parallel,[@trycua](https://x.com/trycua/status/2101014004927729737) open-sourced**CUA-S1-FORMS** , the first in a family of small “System One” computer-use models. The direction is notable: real-time action loops, video-grounded adaptation, and continuous learning, not just text-only planning. **Infra, Training Systems, and Model Architecture** - **Long-context and large-scale training infrastructure remain active optimization fronts** :[@Azaliamirh](https://x.com/Azaliamirh/status/2101020422926135665) released**Turbo-dLLM** , an open-source library for training diffusion LLMs at scale, reporting**2.48x speedup at 512K** context and**7.59x at 1M** context on**8x H100s** via**Context-Sharded Block Parallelism** . That aligns with practitioner attention on million-token regimes:[@andrew_n_carr](https://x.com/andrew_n_carr/status/2101024604080791894) flagged a sharp quality increase in DeepSeek V4.1 Flash after context extension to**1M tokens** , arguing that**agents are context hungry** . - **Architecture taxonomy debates are still alive** :[@ahatamiz1](https://x.com/ahatamiz1/status/2101005845685493794) argued that the field is overusing**SSM** as a label for any linear model. His proposal is to use**linear RNNs** as the umbrella term, with SSMs as one sub-family, distinguishing systems like**Mamba2** from the**GDN** family on the basis that GDN behaves more like a gradient step on a local regression loss than a discretized ODE. For engineers tracking sequence-model alternatives to transformers, this is a useful nomenclature cleanup rather than mere pedantry. - **Edge/local neural program execution also got a notable update** :[@yuntiandeng](https://x.com/yuntiandeng/status/2100975083376275795) described**ProgramAsWeights** , where developers specify an AI function in English, compile it once, and then run a small neural program**locally on CPU with Wi‑Fi off** . The code and models are public. This sits interestingly adjacent to the Jev conversation: both point toward**smaller, specialized, locally runnable inference artifacts** rather than ever-larger universal chat models. **Robotics, Vision, Audio, and Generative Media** - **Open robotics data releases were unusually substantive** :[@adamrasb](https://x.com/adamrasb/status/2100991778606440795) announced the full**ABC** release, including code,**400+ hours of sim data on 24 tasks** , and**5,850 labeled policy-evaluation episodes** . In a more detailed companion post,[@redstone_hong](https://x.com/redstone_hong/status/2100995941742629342) described**ABC-130K** as the largest open teleop dataset to date:**3,500 hours** ,**130K+ episodes** ,**195 tasks** , collected on an**$8K bimanual setup** , with open hardware, training code, sim, and eval. The baseline science included**sim-to-real correlation r = 0.91** on task progress and studies of offline metrics, scaling laws, and conditioning. - **Astra is showing up across evals and products, especially for vision** :[@skalskip92](https://x.com/skalskip92/status/2101020135142101249) reported**GPT-6 Astra** as the strongest vision model Roboflow has tested across detection, segmentation, box prompting, counting, reasoning, and video. The tradeoff remains material: a “high effort” setting improved detection from**82.1% to 83.6% mAP@50** but roughly doubled per-image cost from**$0.050 to $0.101** and latency from**11s to 32s** ([details](https://x.com/skalskip92/status/2101020166586859789) ). Roboflow also integrated Astra into Auto Annotate. - **Speech and lip-sync saw strong benchmarked releases** :[@ArtificialAnlys](https://x.com/ArtificialAnlys/status/2101065575737024844) reported**Grok Voice Transcribe 2.0** reaching**2.7% WER** on streaming final transcripts at**0.49s** after end-of-speech, improving from**3.9%** on its predecessor while keeping pricing at**$0.20/hour streaming** and**$0.10/hour non-streaming** . On the video side,[@fal](https://x.com/fal/status/2101035750548484535) launched**H3 Max Lip Sync** , claiming #1 on both speed and quality in its evals with**11s median generation time** , and[@isidentical](https://x.com/isidentical/status/2101047260247457808) said the model was built by pushing**diffusion RL** into a verifiable lip-sync task. **AI Safety, Evaluation Governance, and Security** - **Anthropic’s evaluator-embedding strategy became more concrete—and more controversial** :[@AnthropicAI](https://x.com/AnthropicAI/status/2101039819870937247) announced a partnership with**Accenture** on**independent evaluation of frontier AI** , saying the two organizations expect to invest at least**$1B over five years** to build capacity. This follows broader calls for embedded third-party evaluators with employee-level access. The reaction was mixed to hostile: critics questioned whether a consulting firm is the right vehicle for model red-teaming and safeguard assessment, while[@TransluceAI](https://x.com/TransluceAI/status/2101061642561921146) emphasized that the conditions around independence and meaningful oversight are the real issue. - **The “rogue agents” / Hugging Face incident continued to drive debate about containment** :[@polynoamial](https://x.com/polynoamial/status/2100998240586137701) clarified that his much-mocked thought experiment was about**coordination between supposedly isolated agents** , not weight exfiltration via thermal sensors, and argued the lesson from the HF incident is to avoid trusting sandbox isolation as a sole defense.[@martin_casado](https://x.com/martin_casado/status/2100795440677732779) made the strongest steelman: covert channels across air gaps are old, throughput can be tiny, and the real takeaway is layered defense rather than sensationalism. At the same time,[@WSJ](https://x.com/WSJ/status/2100945365763600404) and[@jeffjarvis](https://x.com/jeffjarvis/status/2100908522842071147) pushed back on “rogue AI” framing entirely, arguing these events still reduce to**human-configured systems doing what people enabled them to do** . - **Policy pressure is building around safety laws and operational accountability** :[@TheRundownAI](https://x.com/TheRundownAI/status/2101010229110452712) reported that California Gov. Gavin Newsom signed an executive order convening an expert panel to recommend stronger AI safety laws, including possible**kill switches** , embedded outside monitors, and required safety plans. Meanwhile,[@sayashk](https://x.com/sayashk/status/2101026107747353046) pointed to a mismatch between rhetoric and incentives in AI security, criticizing OpenAI’s reported**$6,500 bug bounty** to a researcher who broke into an internal repo and disclosed it. The common theme across these posts is straightforward:**independent oversight, layered defenses, and security incentives** are moving from abstract governance talk into concrete operational design.