---
title: "Consonance · CARD NO. 2026-11-11"
description: "Anthropic releases Claude Fable 5.2 model lineup — A radar over AI and agents. Every morning, and only what several independent sources are saying at once."
canonical: "https://radar.consonance.fyi/en/jour/2026-11-11"
lang: "en"
published: "2026-11-11"
updated: "2026-11-11T08:46:47.362Z"
---

# Consonance · CARD NO. 2026-11-11

> A radar over AI and agents. Every morning, and only what several independent sources are saying at once.

## The day in brief

_written by the model_

1. **Anthropic releases Claude Fable 5.2 model lineup** — Anthropic has announced Claude Fable 5.2 and Claude Mythos 5.2, while extending distillation prevention to API accounts and rolling out Claude Whiteboards, Sheets, and Data features. (→ 03, 05)
2. **OpenAI introduces GPT-Image-2.6 image generation models** — OpenAI integrated GPT-Image-2.6 image models including Flare and Sunburst into Codex and the API, while bringing business features to ChatGPT. (→ 08, 13, 29)

## Video of the day

**Google releases Gemini 4 Carbon and Gemini Omni 1.2 Flash video generation model** — Google has introduced the Gemini 4 Carbon model designed for long-horizon tasks and multi-step reasoning, alongside the Gemini Omni 1.2 Flash video generation model offering improved cost efficiency and visual consistency.

[Watch the video](https://radar.consonance.fyi/media/2026-11-11/en.mp4?v=13476523) · 1:14 · GENERATED AUTOMATICALLY

- 0:00 Opening
- 0:08 Gemini 4 Carbon
- 0:24 GPT-Image-2.6
- 0:39 Claude Code
- 0:53 InfraRSI: Harnessing Recursive Self-Improvement for Training
- 1:05 Closing

## Being discussed

_Subjects at least three independent accounts raised over the last seven days._

### 01. [Anthropic updates Model Context Protocol and adds embedded app tools to Claude](https://radar.consonance.fyi/en/sujet/4149.md)

_agents · 11 independent accounts · 48 posts · 1 articles · 4 labs · 69,029 interactions_

Anthropic has expanded the capabilities of the Model Context Protocol and app extensions within Claude. Users can now run full applications and utilize interactive connectors directly inside the chat interface.

model context protocol

voices: @AndrewBolis @ClaudeDevs @GithubProjects @aiedge_ @alex_prompter @claudeai @dair_ai @dani_avila7

### 02. [Google releases Gemini 4 Carbon and Gemini Omni 1.2 Flash video generation model](https://radar.consonance.fyi/en/sujet/4150.md)

_multimodal · 16 independent accounts · 120 posts · 5 labs · 218,228 interactions_

Google has introduced the Gemini 4 Carbon model designed for long-horizon tasks and multi-step reasoning, alongside the Gemini Omni 1.2 Flash video generation model offering improved cost efficiency and visual consistency.

google · gemini · gemini 3.8

voices: @AndrewCurran_ @AravSrinivas @ArtificialAnlys @CuiMao @GeminiApp @Google @GoogleAIStudio @GoogleDeepMind

Same story: [Google DeepMind announces Gemini 4 Carbon model and new TPU systems](https://radar.consonance.fyi/en/sujet/4160.md) (6 voices) · [Google globally releases the Gemini 4 Carbon reasoning and science model](https://radar.consonance.fyi/en/sujet/4162.md) (4 voices)

### 03. [Anthropic launches Claude Fable 5.2 and Claude Mythos 5.2 with lower pricing](https://radar.consonance.fyi/en/sujet/4151.md)

_architecture · 15 independent accounts · 34 posts · 3 articles · 5 labs · 427,547 interactions_

Anthropic has announced Claude Fable 5.2 and Claude Mythos 5.2 to handle complex technical challenges. Pricing for Fable 5.2 has been reduced to $8 per million input tokens and $40 per million output tokens.

voices: @AlexFinn @AndrewCurran_ @AnthropicAI @ClaudeDevs @CuiMao @addyosmani @aiedge_ @alex_prompter

Same story: [Anthropic releases Claude Fable 5.2 and details new dreaming memory mechanism](https://radar.consonance.fyi/en/sujet/4165.md) (8 voices) · [Anthropic updates Fable 5.2 and extends distillation prevention to all API accounts](https://radar.consonance.fyi/en/sujet/4174.md) (3 voices)

### 04. [OpenAI lets businesses claim profiles and build Business Agents in ChatGPT](https://radar.consonance.fyi/en/sujet/4152.md)

_agents · 5 independent accounts · 5 posts · 2 articles · 3 labs · 82,376 interactions_

OpenAI has rolled out a feature allowing businesses to claim profiles, manage information, and build custom business agents within ChatGPT to engage with customers and answer inquiries.

voices: @OpenAI @gdb @godofprompt @gregisenberg @thsottiaux

### 05. [Anthropic rolls out Claude Whiteboards, Claude Sheets, and Claude Data features](https://radar.consonance.fyi/en/sujet/4153.md)

_system_design · 4 independent accounts · 6 posts · 1 labs · 158,024 interactions_

Anthropic has begun rolling out new workspace tools for Claude, including whiteboards, spreadsheets, and data analytics features designed to help users plan tasks and organize information.

voices: @Hesamation @MatthewBerman @alex_prompter @claudeai @dani_avila7 @minchoi

### 06. [xAI implements fixes and resets weekly usage limits for Grok Bot](https://radar.consonance.fyi/en/sujet/4154.md)

_system_design · 8 independent accounts · 64 posts · 2 labs · 310,404 interactions_

xAI has initiated fixes to improve token efficiency and behavioral consistency for Grok Bot, while also resetting weekly usage limits across all user accounts.

grok · sglang

voices: @AlexFinn @AndrewCurran_ @CuiMao @GithubProjects @Hesamation @MatthewBerman @TheRundownAI @aiedge_

### 07. [Meta AI launches Muse model lineup featuring Ember, Spark, and Wonder](https://radar.consonance.fyi/en/sujet/4159.md)

_architecture · 8 independent accounts · 57 posts · 3 labs · 91,452 interactions_

Meta AI released the Muse model lineup, introducing the compact Muse Ember alongside updates bringing Active Memory and Active Focus to its broader ecosystem. These specialized models are optimized for repetitive agent tasks such as tool calls and coding subtasks.

muse spark · meta ai

voices: @AIatMeta @AndrewCurran_ @ArtificialAnlys @alliekmiller @arena @dani_avila7 @dotey @finkd

### 08. [OpenAI releases GPT-Image-2.6 models and integrates them into Codex](https://radar.consonance.fyi/en/sujet/4161.md)

_agents · 14 independent accounts · 64 posts · 2 articles · 3 labs · 49,541 interactions_

OpenAI brought the GPT-Image-2.6 models, including Flare and Sunburst, to Codex and the API. These new image generation models operate 50% faster than their predecessors while retaining finer visual details at medium and high quality settings.

codex

voices: @AlexFinn @AndrewCurran_ @ArtificialAnlys @CuiMao @GithubProjects @OpenAIDevs @alex_prompter @c_valenzuelab

### 09. [Anthropic upgrades Claude Code with cloud session management features](https://radar.consonance.fyi/en/sujet/4163.md)

_agents · 12 independent accounts · 59 posts · 1 articles · 4 labs · 65,196 interactions_

Anthropic streamlined repository execution within cloud sessions for Claude Code Pro, Max, and Ultra users. The tool automatically detects required dependencies, network access, and API keys before saving the environment configuration for future sessions.

claude code

voices: @ClaudeDevs @GithubProjects @Hesamation @addyosmani @aiedge_ @alex_prompter @alliekmiller @amorriscode

### 10. [Claude Code optimizes background subagents and Artificial Analysis updates index](https://radar.consonance.fyi/en/sujet/4164.md)

_agents · 4 independent accounts · 23 posts · 1 labs · 106,620 interactions_

Anthropic integrated Haiku 5.5 into background subagent status updates within Claude Code, reducing usage limit consumption by roughly 5%. Simultaneously, Artificial Analysis released version 2 of its Decision Index featuring updated benchmarks and calibration metrics.

claude haiku

voices: @AndrewCurran_ @ArtificialAnlys @ClaudeDevs @TheRundownAI @dair_ai @kimmonismus @minchoi @teortaxesTex

### 11. [OpenAI Releases GPT-6 Cyber Security-Specialized Model](https://radar.consonance.fyi/en/sujet/4167.md)

_architecture · 11 independent accounts · 67 posts · 2 articles · 4 labs · 135,571 interactions_

OpenAI introduced a cyber-security-specialized model line named GPT-6 Cyber, combining advanced system capabilities with model alignment mechanisms. The model has been applied to Linux kernel research and vulnerability detection.

gpt-6

voices: @AndrewCurran_ @Hesamation @OpenAI @TheRundownAI @aiedge_ @alex_prompter @arena @dair_ai

### 12. [Community Uploads Pixel Art and 3D Humanoid Datasets to Hugging Face](https://radar.consonance.fyi/en/sujet/4168.md)

_training · 6 independent accounts · 37 posts · 4 articles · 2 labs · 23,097 interactions_

Multiple digital asset datasets featuring millions of pixel art sprites and tens of thousands of rigged humanoid meshes with skeletons were released on Hugging Face. These resources support research in automated rigging and rendering.

hugging face

voices: @AndrewCurran_ @ClementDelangue @Hesamation @NVIDIAAI @arankomatsuzaki @dair_ai @huggingface @teortaxesTex

### 13. [OpenAI Retires Legacy ChatGPT Voice Modes and Introduces Business Features](https://radar.consonance.fyi/en/sujet/4171.md)

_multimodal · 6 independent accounts · 60 posts · 2 labs · 30,435 interactions_

OpenAI announced the retirement of standard and advanced voice modes from ChatGPT starting December 11, 2026, replaced by the GPT-Live conversational model. Additionally, the platform added business tools including GIF generation for images and verified Business Agents.

chatgpt

voices: @AndrewBolis @Hesamation @OpenAI @OpenAIDevs @aiedge_ @alex_prompter @dotey @godofprompt

### 14. [Google Introduces Long-Horizon Coding and Swarm Features in Antigravity](https://radar.consonance.fyi/en/sujet/4172.md)

_agents · 3 independent accounts · 5 posts · 1 articles · 1 labs · 10,184 interactions_

Google demonstrated long-horizon coding capabilities and the swarm feature using the Antigravity system and Gemini. The technology enables subagents to automate 3D asset modeling, WebGL shader programming, and procedural audio generation.

voices: @Google @simonw @sundarpichai

### 15. [Anthropic Launches Personal Finance Feature for Claude in the US](https://radar.consonance.fyi/en/sujet/4173.md)

_agents · 3 independent accounts · 9 posts · 1 articles · 1 labs · 170,570 interactions_

Anthropic released a beta personal finance feature for Claude users in the US. The system connects to user bank accounts with read-only permissions to answer balance and transaction queries in natural language.

voices: @MatthewBerman @PromptLLM @aiedge_ @claudeai @dotey @kimmonismus @minchoi @rohanpaul_ai

### 16. [Evaluation of Fable 5.2 and Astra reasoning models across benchmarks](https://radar.consonance.fyi/en/sujet/4177.md)

_evaluation · 7 independent accounts · 57 posts · 2 labs · 20,453 interactions_

Reasoning models such as Fable 5.2 record benchmark scores exceeding previous predictive trajectories for the Astra series. Organizations and researchers gain access to Daybreak tools to validate security vulnerabilities.

astra · o-series · ilya sutskever

voices: @AndrewCurran_ @ArtificialAnlys @Hesamation @MatthewBerman @OpenAI @deanwball @emollick @haider1

### 17. [OpenAI adds new leadership to core product management](https://radar.consonance.fyi/en/sujet/4178.md)

_system_design · 4 independent accounts · 12 posts · 2 labs · 91,049 interactions_

A senior product leader previously directing developer tools at Anthropic joins OpenAI to lead core product management. This recruitment strengthens product development and scaling initiatives.

voices: @AndrewCurran_ @CuiMao @_catwu @dani_avila7 @dotey @gdb @jerryjliu0 @kimmonismus

### 18. [Automated cybersecurity pipeline discovers and patches nearly 1,000 Linux vulnerabilities](https://radar.consonance.fyi/en/sujet/4181.md)

_agents · 4 independent accounts · 15 posts · 2 articles · 3 labs · 65,551 interactions_

An automated pipeline leveraging specialized AI models successfully validates and generates patches for 948 local privilege escalation vulnerabilities on Linux. The system also discovers at least 10 previously undocumented exploitation techniques.

voices: @Hesamation @OpenAI @gdb @haider1 @kimmonismus @rohanpaul_ai @scaling01 @thsottiaux

### 19. [Cyber at the Frontier livestream announces new security updates](https://radar.consonance.fyi/en/sujet/4182.md)

_system_design · 4 independent accounts · 5 posts · 1 articles · 3 labs · 11,697 interactions_

New security tools derived from operational defense experiences are introduced during the Cyber at the Frontier livestream event. The presentation outlines practical applications of frontier models for customer protection.

voices: @OpenAI @derrickcchoi @gdb @thsottiaux

### 20. [Open-weights model Inkling 2 Small released with Mixture-of-Experts architecture](https://radar.consonance.fyi/en/sujet/4184.md)

_architecture · 5 independent accounts · 10 posts · 1 articles · 32,961 interactions_

The Mixture-of-Experts model Inkling 2 Small launches featuring 594B total parameters and 17B active parameters, along with image input capabilities. Pre-training completed in 68 hours on GB300 hardware using NVFP4 precision.

voices: @MatthewBerman @arena @kimmonismus @natolambert @soumithchintala @swyx @teortaxesTex @thinkymachines

### 21. [Moonshot AI releases open-weights for 2.8-parameter Kimi K3.1 Pro model](https://radar.consonance.fyi/en/sujet/4186.md)

_architecture · 4 independent accounts · 18 posts · 2 articles · 1 labs · 35,221 interactions_

The Kimi K3.1 Pro model weights are published on open platforms, featuring 2.8 trillion total parameters with 104 billion active parameters. The architecture supports coding and financial research tasks, and integrates with Ollama Cloud.

ollama · gpqa

voices: @AndrewCurran_ @CuiMao @GithubProjects @arena @kimmonismus @ollama @rohanpaul_ai @scaling01

### 22. [Anthropic uses Claude to map vault particle structures in bacteria](https://radar.consonance.fyi/en/sujet/4188.md)

_multimodal · 3 independent accounts · 6 posts · 2 articles · 1 labs · 20,082 interactions_

Anthropic's biology research team used Claude to map the 3D structure of vault particles within bacteria and viruses. These hollow, barrel-shaped protein structures were first discovered in human cells in 1986 but have remained largely unexplained.

voices: @AnthropicAI @Hesamation @dotey @rohanpaul_ai @teortaxesTex

### 23. [Evaluating AI reasoning capabilities through social deduction games](https://radar.consonance.fyi/en/sujet/4189.md)

_evaluation · 4 independent accounts · 6 posts · 1 labs · 13,335 interactions_

Researchers tested large language models using social deduction games like Blood on the Clocktower and economic simulation benchmarks such as CrusaderBench. The evaluations demonstrated notable differences in strategy and progress across various model families over time.

voices: @Hesamation @derrickcchoi @polynoamial @teortaxesTex @dwarkesh_sp

### 24. [Recursive Intelligence releases Ripple-65B Preview coding model](https://radar.consonance.fyi/en/sujet/4192.md)

_architecture · 3 independent accounts · 8 posts · 1 articles · 17,278 interactions_

Recursive Intelligence has launched Ripple-65B Preview, a dense language model pretrained from scratch for 32 million dollars. The model sets a new performance standard for U.S. open-weight systems in agentic coding tasks while outperforming significantly larger alternatives.

voices: @iScienceLuvr @kimmonismus @natolambert @swyx @teortaxesTex @RicursiveAI

### 25. [The dots app updates with inline notifications and fast sharing features](https://radar.consonance.fyi/en/sujet/4193.md)

_agents · 3 independent accounts · 4 posts · 2 labs · 12,515 interactions_

The dots application released a new release focused on performance and reliability improvements. The update introduces inline notification replies on iOS and direct content sharing features.

voices: @ajambrosino @gdb @thsottiaux @_samirism

### 26. [Stripe acquires OpenRouter and expands API integration offerings](https://radar.consonance.fyi/en/sujet/4194.md)

_system_design · 7 independent accounts · 18 posts · 1 labs · 3,444 interactions_

Stripe has acquired the model routing platform OpenRouter to strengthen economic infrastructure for artificial intelligence while maintaining neutral operations. Concurrently, OpenRouter integrated new models including Embed 5, Rerank 5, and Inkling 2 Small under a unified billing system.

openrouter

voices: @ArtificialAnlys @OpenRouter @cohere @godofprompt @natolambert @soumithchintala @teortaxesTex @testingcatalog

### 27. [The dots team recruits engineers to expand its persistent agent ecosystem](https://radar.consonance.fyi/en/sujet/4196.md)

_agents · 3 independent accounts · 5 posts · 1 labs · 7,351 interactions_

The team behind dots is actively recruiting engineers and specialists in the persistent agent space. The hiring drive aims to assemble technical talent to scale product development.

voices: @ajambrosino @dotey @edbayes @thsottiaux @_samirism

### 28. [Cognition launches SWE-2 Flash built on the Inkling-small base model](https://radar.consonance.fyi/en/sujet/4198.md)

_training · 3 independent accounts · 5 posts · 10,629 interactions_

Cognition introduced SWE-2 Flash, an optimized coding model post-trained on top of the Inkling-small base model. The release matches larger frontier benchmarks while delivering three times faster token generation and five times lower costs.

voices: @cognition @kimmonismus @soumithchintala @swyx @teortaxesTex

### 29. [OpenAI releases Images 2.6 featuring GPT-Image-2.6 Flare and Sunburst](https://radar.consonance.fyi/en/sujet/4199.md)

_multimodal · 3 independent accounts · 6 posts · 2 articles · 2 labs · 7,188 interactions_

OpenAI has introduced Images 2.6, featuring the new GPT-Image-2.6 Flare and Sunburst models. Flare generates images in as little as six seconds, while Sunburst provides higher quality and greater consistency across edits. Both models are available via the API and integrated into ChatGPT and Codex.

voices: @OpenAI @OpenAIDevs @arena @TownAI

### 30. [AI community debates the limits of compute scaling and the bitter lesson](https://radar.consonance.fyi/en/sujet/4202.md)

_evaluation · 3 independent accounts · 5 posts · 4,280 interactions_

Researchers and developers are discussing the role of data and compute scaling in the generalization power of artificial intelligence. The debate centers on whether brute-force computation alone is sufficient for intelligence or if new scientific methods are required.

voices: @Hesamation @fchollet @giffmana @RichardSSutton

## Worth reading closely

_Papers and writeups, read from their abstracts, ranked by relevance to someone building agents and backends._

### 01. [InfraRSI: Harnessing Recursive Self-Improvement for Training Infrastructure](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.11558.md)

_agents · 33 upvotes_

**QUESTION** — How can distributed LLM training optimization be automated via a recursive self-improvement loop while avoiding out-of-memory errors or numerical drift?

- Across four single- and multi-node dense and Mixture-of-Experts (MoE) LLM workloads, InfraRSI achieves a speedup of up to 2.75times.

yuchuantian · 8 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.11558)

### 02. [World In-Context Learning for Agentic Robot Manipulation](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.12885.md)

_agents · 57 upvotes_

**QUESTION** — How can robots learn new manipulation tasks in-context without updating network weights?

- RoboScript reaches 83.4% mean success across all 50 tasks of RoboTwin 2.0 with one demonstration per task and no gradient updates.

dkliang · 9 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.12885)

### 03. [CPU-ARC: Agentic Reasoning Caching for on-CPU Reasoning Reuse](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.09891.md)

_agents · 10 upvotes_

**QUESTION** — How can LLM reasoning be reused efficiently for new problem instances to reduce costs without invoking the original language models?

- ARCs show on-par performance with frontier LLMs across a range of difficult mathematical and scientific tasks.

pbelcak · 6 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.09891)

### 04. [Infrathon: Benchmarking Language Agents for Real-World, Long-Horizon AI Infra Optimization](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.11995.md)

_evaluation · 52 upvotes_

**QUESTION** — Can frontier AI agents continually improve real-world AI infrastructure over long horizons?

- 135 of the 150 runs achieve further improvements during the second half of their budgets.

TTTXXX01 · 9 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.11995)

### 05. [How Much Training-Inference Mismatch Can On-Policy Distillation Tolerate?](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.11450.md)

_training · 13 upvotes_

**QUESTION** — How much training-inference mismatch can on-policy distillation tolerate before suffering degradation?

- OPD tolerates substantial mismatch without meaningful loss of training quality but degrades at higher levels, revealing an empirical mismatch budget.

Benyucong · 8 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.11450)

### 06. [Authorization Drift in LLM Agents: Evaluating Boundary Compliance in Multi-Step Tool Use](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.09113.md)

_agents · 7 upvotes_

**QUESTION** — How frequently and in what ways do LLM agents exceed their authorization boundaries during multi-step tool use?

- Across 13 LLMs and three frameworks (25,207 trajectories), greater capability accompanies greater drift, with frontier-capable models exhibiting drift rates of 0.43 to 0.61.

chuxuecao · 6 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.09113)

### 07. [Reframe: Observation-Aligned Multi-Teacher On-Policy Distillation for Agentic Visual Reasoning](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.12886.md)

_training · 40 upvotes_

**QUESTION** — How can we improve post-training for tool-using visual agents without overlooking returned observations?

- Reframe outperforms both Mix-RL and vanilla MOPD over 14 image and video benchmarks.

EliYuan00 · 9 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.12886)

### 08. [How Many Exposures Does a World Model Need? Learning Unseen Game Mechanics at Test Time](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.12324.md)

_agents · 15 upvotes_

**QUESTION** — How many exposures does a world model need to predict the effects of a new environment's mechanics at test time?

- WARC learns to predict four in five mechanics with at least 80% accuracy on the cells they change and keeps that accuracy to the end of the recordings.

KuOnoda · 9 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.12324)

### 09. [TRIO: Task Refinement with Individual Optimization for Retrieval-Augmented Generation](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.02810.md)

_agents · 8 upvotes_

**QUESTION** — How can multiple AI agents be coordinated within a RAG system to minimize hallucinations and improve retrieval quality?

- TRIO equips three well-designed agents with an iterative refinement paradigm by aligning their individual optimization objectives.

An998 · 2 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.02810)

### 10. [PhysOPSD: Code as Privilege for Physically Plausible Video Generation](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.12411.md)

_training · 8 upvotes_

**QUESTION** — How can physical plausibility in video generation be improved without sacrificing overall visual fidelity?

- PhysOPSD introduces executable code as privileged physical context for video generation, instantiating an editable physical world with intrinsic abstraction and controllability.

Harold328 · 9 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.12411)

### 11. [Learning Heterogeneous Tactile Representation for General Dexterous Manipulation](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.12806.md)

_multimodal · 99 upvotes_

**QUESTION** — How can we construct a unified tactile representation that integrates across different sensor types for dexterous manipulation models?

- The model yields a unified representation across piezoresistive, magnetic, and visuotactile sensors.

cherubicxn · 9 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.12806)

### 12. [MMDM: Measuring Massive Multitask Decision Making](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.14309.md)

_evaluation · 28 upvotes_

**QUESTION** — How can we evaluate the complex multi-task decision-making capabilities of System-1 Decision Models?

- MMDM consists of 5,448 questions across 16 task categories with option spaces ranging from 2 to 1,024 and inputs extending to 64,817 tokens.

shenjunhao · 10 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.14309)

### 13. [DexForesight: Grounding Dexterous Policies with Implicit Spatial Foresight](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.11166.md)

_multimodal · 26 upvotes_

**QUESTION** — How can spatial dynamics knowledge be injected into the latent space of a dexterous policy without requiring explicit future prediction at deployment?

- On 11 DexJoCo simulation tasks, DexForesight improves the overall success rate of the base policy from 49.4% to 60.5%.

ChenZhongxi · 10 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.11166)

### 14. [SWAM: Scaling World Action Models with Cross-Modal Alignment](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.12870.md)

_multimodal · 23 upvotes_

**QUESTION** — How can language-video alignment be maintained and improved when incorporating action learning into world action models?

- The authors curate a pool of approximately 47,000 hours of general videos, egocentric videos, UMI demonstrations, and robot demonstrations.

FreDERicia · 9 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.12870)

### 15. [DepthGraft: Structural Regularization through Hierarchical Cross-Layer KV Reconstruction](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.03142.md)

_architecture · 15 upvotes_

**QUESTION** — How can cross-layer KV sharing serve as an implicit structural regularizer to improve deep-layer utilization in language models?

- DepthGraft distributes non-local gradient flow across source layers, while preserving prefill early completion and reducing KV cache by roughly one third.

galaxy-1 · 2 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.03142)

### 16. [FewStepRL: Fine-tuning Few-step Diffusion Models with Reinforcement Learning](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.08240.md)

_training · 9 upvotes_

**QUESTION** — How can reinforcement learning be systematically adapted to few-step diffusion models without incurring prohibitive rollout costs?

- The framework integrates FlowGRPO, AWM, DiffusionNFT, DiffusionOPSD, and ReFL into one modular codebase.

Delcher · 5 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.08240)

### 17. [dNaViT: Discrete Codes are General-Purpose Visual Representations](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.12426.md)

_architecture · 35 upvotes_

**QUESTION** — How far can a finite discrete vocabulary go in representing the richness of the visual world?

- dNaViT matches or exceeds task-specific baselines across 12 visual understanding benchmarks, 4 generation benchmarks, and an image editing benchmark.

XiaotongLi97 · 9 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.12426)

### 18. [ProTEUS: A mutation-resolved world model for closed-loop protein evolution](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.13348.md)

_multimodal · 14 upvotes_

**QUESTION** — How can the functional consequences of protein mutations be accurately modeled and predicted for closed-loop protein evolution?

- ProTEUS learns a joint representation of sequence, evolutionary, and structural information using MutationFormer, a mutation-biased attention architecture.

yikuuun · 9 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.13348)

### 19. [Few-Step Streaming Video Generation via Condition Shifting](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.14033.md)

_multimodal · 88 upvotes_

**QUESTION** — How can condition shifting be used to eliminate the independent fake network in few-step streaming video generation?

- Compared with S2-Avatar, APEX achieves higher video instruction fulfillment on the Progressive track (3.856 versus 3.831) and higher visual aesthetics across both tracks (0.641898 versus 0.632606).

SempraETY · 10 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.14033)

### 20. [MotionGraphicsBench: Evaluating Motion Graphics Generation with Programmatic Verification](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.10592.md)

_evaluation · 14 upvotes_

**QUESTION** — How can motion graphics generation by language models be evaluated without relying on VLM judges or reference animations?

- On a balanced test set of 180 animations, our programs achieve 93.3% raw agreement with human judgments, compared with 83.3% for a GPT-6 Astra VLM baseline, and outperform the baseline in both approval and rejection recall.

jama1017 · 7 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.10592)

### 21. [Longtail-SWE: High-Quality SWE Tasks for Underrepresented Languages](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.10304.md)

_evaluation · 22 upvotes_

**QUESTION** — How can high-quality repository-level benchmarks be constructed for programming languages underrepresented in existing datasets?

- Longtail-SWE is a benchmark of 202 issue-resolution tasks across 13 languages with limited coverage in existing repository-level benchmarks.

djalexj · 6 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.10304)

### 22. [All Quiet on the Neural Front: Language Models Do Not Report Injected Pain States](https://radar.consonance.fyi/en/lecture/arxiv%3A2611.10123.md)

_evaluation · 7 upvotes_

**QUESTION** — Can language models verbally report the pain states they internally represent and execute upon when steered?

- Across 52,920 reports from 7B and 8B models spanning 120 tasks, four doses and three prompts, pain reports never exceeded what altered text revealed.

emilioferrara · 6 November 2026 · [read the original ↗](https://arxiv.org/abs/2611.10123)

## Hands-on

_Repositories climbing on GitHub today, scored on the day's momentum, rank, adoption, freshness and project health._

- [meta/astryx](https://github.com/meta/astryx): An open-source design system built specifically for agentic workflows and automated UI generation. · +3,108 stars today · ★ 23,803 · TypeScript
- [anthropics/skills](https://github.com/anthropics/skills): Official Anthropic toolkit providing reusable skills and capabilities for developers building autonomous agent architectures. · +1,679 stars today · ★ 192,687 · Python
- [addyosmani/agent-skills](https://github.com/addyosmani/agent-skills): Provides production-grade engineering skills for developers building autonomous coding agents. · +465 stars today · ★ 115,434 · JavaScript
- [anthropics/claude-quickstarts](https://github.com/anthropics/claude-quickstarts): This collection of reference implementations helps backend engineers quickly integrate LLM APIs into production applications. · +531 stars today · ★ 20,662 · Python
- [google-deepmind/mujoco](https://github.com/google-deepmind/mujoco): This high-performance physics simulator is essential for training robotics models and physical world environments for agents. · +456 stars today · ★ 17,522 · C++
- [vercel-labs/agent-browser](https://github.com/vercel-labs/agent-browser): A Rust-based browser automation CLI designed for AI agents to interact with web interfaces reliably. · +780 stars today · ★ 49,484 · Rust
- [homeassistant-ai/ha-mcp](https://github.com/homeassistant-ai/ha-mcp): This Model Context Protocol integration allows AI assistants and backend agents to directly control smart home hardware. · +790 stars today · ★ 6,872 · Python
- [mattpocock/skills](https://github.com/mattpocock/skills): A curated collection of prompt engineering skills and shell configurations for engineers optimizing coding agent workflows. · +1,225 stars today · ★ 321,326 · Shell
- [ranxianglei/billion-context](https://github.com/ranxianglei/billion-context): A context compression plugin that enables month-long LLM sessions within small windows while reducing token usage. · +375 stars today · ★ 6,596 · TypeScript

## Claims

_Checkable assertions with their sources and their contradictions. Each stands on at least two independent sources, or a person read it first; the stamp says which._

### 01. Stripe acquired the model routing platform OpenRouter.

_TWO SOURCES · NOT REREAD · 11 November 2026 · openrouter · 2 sources_

- **evidence** · [@stripe](https://x.com/stripe/status/2120240651778355306): “OpenRouter is now part of Stripe.”
- **evidence** · [@testingcatalog](https://x.com/testingcatalog/status/2120245174752137605): “BREAKING 🔥: Stripe acquired OpenRouter!!!”
- **evidence** · [@OpenRouter](https://x.com/OpenRouter/status/2120241577649909899): “OpenRouter is officially part of @stripe today!”

**Still to check**
- What was the deal value and acquisition terms?
- How will the technical integration between Stripe and OpenRouter proceed?

## Next

- [Previous day: 10 November](https://radar.consonance.fyi/en/jour/2026-11-10.md)
- [Next day: 13 November](https://radar.consonance.fyi/en/jour/2026-11-13.md)
- [archives](https://radar.consonance.fyi/en/archives.md)
- [method](https://radar.consonance.fyi/en/methode.md)
- [Consonance](https://radar.consonance.fyi/en/index.md)
- Languages: [Français](https://radar.consonance.fyi/jour/2026-11-11.md) · [Tiếng Việt](https://radar.consonance.fyi/vi/jour/2026-11-11.md)
- For machines: [llms.txt](https://radar.consonance.fyi/llms.txt) · [RSS](https://radar.consonance.fyi/en/rss.xml) · [sitemap.xml](https://radar.consonance.fyi/sitemap.xml)
- HTML version: https://radar.consonance.fyi/en/jour/2026-11-11
