DeepSeek Ties Its New V4.1-Flash-1017 Model Tighter to Its Own Harness
Model tied to DeepSeek's own harness
AI Storyline
45 items · 10 sources · 17 days
Operational story trace
DeepSeek Open-sources
Follow in this browser to see new updates on your Live feed.
Current stateDeveloping
Latest change
Coverage of the dsh-superagent prototype, a router-worker coding agent that watches team chat for work, continued through Nov 4, while the rest of the thread is third-party inference-engine news.
Earlier contextThe story so far
DeepSeek tied its V4.1-Flash model to its own agent harness on Oct 17, then on Oct 20 open-sourced DeepInfer, the inference engine behind its production services. On Oct 31 it published the architecture of an always-on coding agent, dsh-superagent.
editor-curated · source-linked
Arc
Oct 17Nov 6 · now
HARNESS · Oct 17V4.1-Flash-1017 tied tighter to DeepSeek's own harness1 source · show source ▾
DEEPINFER · Oct 20–21DeepSeek open-sources DeepInfer, its production inference engineWide press pickup of the engine, with claims of 155T tokens a day per cluster.10 sources · show sources ▾
DeepSeek Unveils Open-Source Tool to Help Cut AI Operating Costs
DeepSeek Unveils Open-Source Tool to Help Cut AI Operating Costs Bloomberg.com
DeepSeek is open-sourcing the software it uses to run its AI models at scale
DeepSeek is open-sourcing the software it uses to run its AI models at scale Quartz
DeepSeek Dials Up the Pressure on OpenAI and Anthropic With Open-Source Inference Tool
DeepSeek Dials Up the Pressure on OpenAI and Anthropic With Open-Source Inference Tool Gizmodo
DeepSeek Open Sources Inference Tech to Lower AI Costs
DeepSeek Open Sources Inference Tech to Lower AI Costs PYMNTS.com
DeepSeek open-sources DeepInfer, the inference framework behind its production services
Names DeepInfer as the engine behind production services
DeepSeek Opens DeepInfer, Presses OpenAI, Anthropic [2026]
DeepSeek Opens DeepInfer, Presses OpenAI, Anthropic [2026] shattered.io
DeepSeek Open-Sources Framework Behind Its High-Margin AI Inference
DeepSeek Open-Sources Framework Behind Its High-Margin AI Inference The Information
[AINews] DeepSeek DeepInfer: Open Source SOTA Inference running 155T toks/day/cluster making $22m/day at ~80% margin for a 81 day payback
Adds the 155T tokens/day and revenue claims
DeepSeek open-sources DeepInfer, the engine serving 155T tokens a day
DeepSeek open-sources DeepInfer, the engine serving 155T tokens a day Dealroom
DeepSeek Open-Sources DeepInfer, Its Production Inference Engine
DeepSeek Open-Sources DeepInfer, Its Production Inference Engine Technology Org
ADJACENT · Oct 21 – Nov 3Loosely related research and framework papers clustered in17 sources · show sources ▾
VX-One: Typed, Calibrated Decision Heads for Real-Time Perimeter Triage - An empirical study of unknown-attack honesty, cascade gray zones, and calibration under drift
Perimeter triage faces a latency/semantics trade-off: signature engines are fast and format-safe but blind to unseen attacks, while generative models are semantic but slow and produce unparseable output. In late 2026...
Can LLMs Improve Their Own Harness? A Controlled Study on ARC-AGI-3
A generally intelligent system acquires new skills efficiently without requiring humans to redesign its software scaffold for every new task. Yet language models rarely act alone. A harness is a symbolic program that...
Better Reasoning, Not Just Bigger Models: A Structure-Grounded Framework for Efficient and Reliable Inference
Scaling large language models has substantially improved linguistic fluency, yet reliable autonomous agents require more than fluent generation. In multi-turn inference, current agentic systems often produce semantica...
A framework for evaluating agentic AI systems
A framework for evaluating agentic AI systems
Tracing Hallucinations in LLM Reasoning for UML Class Diagram Generation: An Empirical Taxonomy and Mitigation Study
Large Language Models (LLMs) can generate UML class diagrams from natural language requirements, but hallucinations may compromise the correctness and completeness of the resulting diagrams. Existing evaluations mostl...
How Well Do Decision Models Transfer to Vietnamese? An Empirical Study of Text Classification
Recent advances in specialized decision models have expanded the possibilities for zero-shot classification, yet their effectiveness on Vietnamese remains underexplored. This study evaluates four specialized decision...
An Empirical Study of Recent Large Language Models for Knowledge Graph to Text Generation on Standard Benchmarks
Knowledge graph-to-text (KG-to-text) generation verbalizes a set of knowledge graph triples into fluent natural language. While fine-tuned encoder-decoder models have long been the standard approach, recent LLMs are b...
Social-HUMINT: An LLM-Augmented Unified Framework for Measuring Social Media Exposure
In the post-API landscape, social-media profiling has become inexpensive and automated: Large Language Models (LLMs) can extract structured information from unstructured content and deduce attributes users never discl...
Do Language Models Trust Their Own Justifications? A Study on Functional Consistency
Large Language Models (LLMs) have been widely adopted in text classification tasks, where they not only output class predictions but also generate explanations that highlight the tokens deemed most relevant for reachi...
EvolveMAS: A Unified Self-Evolution Framework for Decentralized Multi-Agent Systems
Decentralized multi-agent systems (MAS) enable large language model agents to interact without centralized control. Agents need to adapt to other agents at test time and reuse accumulated experience to refine their in...
CEPO: A General Framework for Improving Prompt Optimization in Multi-Agent Systems
Multi-agent systems (MAS) require both effective coordination and specialized task execution. Existing prompt optimization approaches typically couple these functions within agent-specific prompts. We introduce Coordi...
A Unified Hessian-Guided Framework for Post-Training Quantization
Low-bit quantization enables efficient deployment of large language models by reducing memory footprint and computational cost. Achieving high accuracy at low precision requires selecting quantization ranges, linear transformations, and weight rounding decisions. Existing approac
Confident Fools or Careful Searchers? A Behavioural Study of When LLMs Search Beyond Their Static Internal Memory
Agentic LLM assistants must repeatedly decide whether to answer from memory or to call a tool such as web search. Unnecessary searches cost latency and money; missing ones yield confident but unsupported answers. Prio...
Compression Technique Interactions in Small Language Models: A Factorial Ablation Study
Deploying small language models on resource-constrained hardware requires compression, yet practitioners lack systematic guidance on how quantization, pruning, and post-compression recovery interact when combined. We...
WeSearch: A Unified Model-Centric Retrieval Framework for WeChat Search
We present WeSearch, a unified, model-centric retrieval framework for large-scale, continuously evolving content search. Prevailing methods assign documents fixed hierarchical Semantic IDs, implicitly assuming each do...
An Axiomatic Framework for Reinforcement Learning in Open Multi-Agent Systems
Open multi-agent systems, where only a subset of agents are controllable, are central to real-world domains such as smart grids, yet lack a principled foundation for reinforcement learning algorithm design. We study this setting through n-agent ad hoc teamwork (NAHT), a recent fr
Aligning Language Models with General Preferences: An Empirical Study of Direct Nash Optimization
Direct Nash Optimization (DNO) aligns a language model to a general preference function by approximating the Nash equilibrium of a two-player preference game. Its practical variant, DNO-Prct, iterates on-policy sampling, LLM-based rating of the samples together with a teacher's g
SUPER-AGENT · Oct 31 – Nov 4dsh-superagent: a router-worker coding agent goes open source9 sources · show sources ▾
DeepSeek put the architecture for their always-on autonomous agent on GitHub
First report of the always-on agent architecture on GitHub
DeepSeek Open Sources the Architecture Behind Its Super-Agent
DeepSeek Open Sources the Architecture Behind Its Super-Agent Startup Fortune
DeepSeek Harness Team Open-Sources a Super-agent Prototype That Finds and Fixes Bugs on Its Own
DeepSeek Harness Team Open-Sources a Super-agent Prototype That Finds and Fixes Bugs on Its Own Pandaily
DeepSeek Unveils Open-Source 'Super-Agent' Coding Tool in Direct Challenge to Anthropic's Claude Code
DeepSeek Unveils Open-Source 'Super-Agent' Coding Tool in Direct Challenge to Anthropic's Claude Code BigGo Finance
DeepSeek Open-Sources dsh-superagent, a Router-Worker Coding Agent
Names dsh-superagent and its router-worker design
DeepSeek open-sources dsh-superagent research prototype
DeepSeek open-sources dsh-superagent research prototype TNGlobal
DeepSeek team open-sources proactive coding agent research prototype
DeepSeek team open-sources proactive coding agent research prototype TechNode
DeepSeek open-sources Super-Agent, a coding agent that watches team chats for work
DeepSeek open-sources Super-Agent, a coding agent that watches team chats for work Help Net Security
DeepSeek Open-Sources Super-Agent, a Coding Agent That Works From Team Chat
Latest DeepSeek coverage of the team-chat workflow
ECOSYSTEM · Oct 18 – Nov 6Other inference engines and serving economics in the same window8 sources · show sources ▾
New Co-Design Framework Speeds Up LLM Inference on Mixed Hardware by Taming Memory Bottlenecks
New Co-Design Framework Speeds Up LLM Inference on Mixed Hardware by Taming Memory Bottlenecks Bioengineer.org
llm-d 1.0 Advances Open, Cost-Efficient AI Inference for Production Scale
llm-d 1.0 Advances Open, Cost-Efficient AI Inference for Production Scale TradingView
Why founders keep failing to vibecode their own LLM inference engine
Why founders keep failing to vibecode their own LLM inference engine Startup Fortune
Inference engine in game scripting language using macros
Inference engine in game scripting language using macros
Flare, Nvidia's DGX Spark Inference Engine
Flare, Nvidia's DGX Spark Inference Engine
A Cross-Domain Matrix Engine for Signal Processing and Deep Neural Net Inference
Matrix multiplication dominates both PHY signal processing and Deep Neural Network (DNN) inference, yet these domains place opposing demands on the matrix engine. Signal processing requires low-latency execution of relatively small matrix operations, whereas DNN inference, partic
Is hosting an open-source inference engine profitable?
Is hosting an open-source inference engine profitable?
Supercharging XLA-Based Inference: Announcing the Launch of XLA Serving Engine
Newest item: a separate XLA-based serving engine
Day 1 Saturday, Oct 17, 2026
DeepSeek Ties Its New V4.1-Flash-1017 Model Tighter to Its Own Harness
Model tied to DeepSeek's own harness
Day 2 Sunday, Oct 18, 2026
New Co-Design Framework Speeds Up LLM Inference on Mixed Hardware by Taming Memory Bottlenecks
New Co-Design Framework Speeds Up LLM Inference on Mixed Hardware by Taming Memory Bottlenecks Bioengineer.org
Day 3 Tuesday, Oct 20, 2026
llm-d 1.0 Advances Open, Cost-Efficient AI Inference for Production Scale
llm-d 1.0 Advances Open, Cost-Efficient AI Inference for Production Scale TradingView
DeepSeek Unveils Open-Source Tool to Help Cut AI Operating Costs
DeepSeek Unveils Open-Source Tool to Help Cut AI Operating Costs Bloomberg.com
DeepSeek is open-sourcing the software it uses to run its AI models at scale
DeepSeek is open-sourcing the software it uses to run its AI models at scale Quartz
DeepSeek Dials Up the Pressure on OpenAI and Anthropic With Open-Source Inference Tool
DeepSeek Dials Up the Pressure on OpenAI and Anthropic With Open-Source Inference Tool Gizmodo
DeepSeek Open Sources Inference Tech to Lower AI Costs
DeepSeek Open Sources Inference Tech to Lower AI Costs PYMNTS.com
DeepSeek open-sources DeepInfer, the inference framework behind its production services
Names DeepInfer as the engine behind production services
Day 4 Wednesday, Oct 21, 2026
DeepSeek Opens DeepInfer, Presses OpenAI, Anthropic [2026]
DeepSeek Opens DeepInfer, Presses OpenAI, Anthropic [2026] shattered.io
DeepSeek Open-Sources Framework Behind Its High-Margin AI Inference
DeepSeek Open-Sources Framework Behind Its High-Margin AI Inference The Information
[AINews] DeepSeek DeepInfer: Open Source SOTA Inference running 155T toks/day/cluster making $22m/day at ~80% margin for a 81 day payback
Adds the 155T tokens/day and revenue claims
DeepSeek open-sources DeepInfer, the engine serving 155T tokens a day
DeepSeek open-sources DeepInfer, the engine serving 155T tokens a day Dealroom
DeepSeek Open-Sources DeepInfer, Its Production Inference Engine
DeepSeek Open-Sources DeepInfer, Its Production Inference Engine Technology Org
VX-One: Typed, Calibrated Decision Heads for Real-Time Perimeter Triage - An empirical study of unknown-attack honesty, cascade gray zones, and calibration under drift
Perimeter triage faces a latency/semantics trade-off: signature engines are fast and format-safe but blind to unseen attacks, while generative models are semantic but slow and produce unparseable output. In late 2026...
Can LLMs Improve Their Own Harness? A Controlled Study on ARC-AGI-3
A generally intelligent system acquires new skills efficiently without requiring humans to redesign its software scaffold for every new task. Yet language models rarely act alone. A harness is a symbolic program that...
Day 5 Thursday, Oct 22, 2026
Better Reasoning, Not Just Bigger Models: A Structure-Grounded Framework for Efficient and Reliable Inference
Scaling large language models has substantially improved linguistic fluency, yet reliable autonomous agents require more than fluent generation. In multi-turn inference, current agentic systems often produce semantica...
A framework for evaluating agentic AI systems
A framework for evaluating agentic AI systems
Tracing Hallucinations in LLM Reasoning for UML Class Diagram Generation: An Empirical Taxonomy and Mitigation Study
Large Language Models (LLMs) can generate UML class diagrams from natural language requirements, but hallucinations may compromise the correctness and completeness of the resulting diagrams. Existing evaluations mostl...
How Well Do Decision Models Transfer to Vietnamese? An Empirical Study of Text Classification
Recent advances in specialized decision models have expanded the possibilities for zero-shot classification, yet their effectiveness on Vietnamese remains underexplored. This study evaluates four specialized decision...
An Empirical Study of Recent Large Language Models for Knowledge Graph to Text Generation on Standard Benchmarks
Knowledge graph-to-text (KG-to-text) generation verbalizes a set of knowledge graph triples into fluent natural language. While fine-tuned encoder-decoder models have long been the standard approach, recent LLMs are b...
Social-HUMINT: An LLM-Augmented Unified Framework for Measuring Social Media Exposure
In the post-API landscape, social-media profiling has become inexpensive and automated: Large Language Models (LLMs) can extract structured information from unstructured content and deduce attributes users never discl...
Day 6 Friday, Oct 23, 2026
Do Language Models Trust Their Own Justifications? A Study on Functional Consistency
Large Language Models (LLMs) have been widely adopted in text classification tasks, where they not only output class predictions but also generate explanations that highlight the tokens deemed most relevant for reachi...
Day 7 Saturday, Oct 24, 2026
Why founders keep failing to vibecode their own LLM inference engine
Why founders keep failing to vibecode their own LLM inference engine Startup Fortune
Day 8 Sunday, Oct 25, 2026
EvolveMAS: A Unified Self-Evolution Framework for Decentralized Multi-Agent Systems
Decentralized multi-agent systems (MAS) enable large language model agents to interact without centralized control. Agents need to adapt to other agents at test time and reuse accumulated experience to refine their in...
Day 9 Monday, Oct 26, 2026
CEPO: A General Framework for Improving Prompt Optimization in Multi-Agent Systems
Multi-agent systems (MAS) require both effective coordination and specialized task execution. Existing prompt optimization approaches typically couple these functions within agent-specific prompts. We introduce Coordi...
Day 10 Thursday, Oct 29, 2026
A Unified Hessian-Guided Framework for Post-Training Quantization
Low-bit quantization enables efficient deployment of large language models by reducing memory footprint and computational cost. Achieving high accuracy at low precision requires selecting quantization ranges, linear transformations, and weight rounding decisions. Existing approac
Day 11 Friday, Oct 30, 2026
Confident Fools or Careful Searchers? A Behavioural Study of When LLMs Search Beyond Their Static Internal Memory
Agentic LLM assistants must repeatedly decide whether to answer from memory or to call a tool such as web search. Unnecessary searches cost latency and money; missing ones yield confident but unsupported answers. Prio...
Compression Technique Interactions in Small Language Models: A Factorial Ablation Study
Deploying small language models on resource-constrained hardware requires compression, yet practitioners lack systematic guidance on how quantization, pruning, and post-compression recovery interact when combined. We...
Day 12 Saturday, Oct 31, 2026
WeSearch: A Unified Model-Centric Retrieval Framework for WeChat Search
We present WeSearch, a unified, model-centric retrieval framework for large-scale, continuously evolving content search. Prevailing methods assign documents fixed hierarchical Semantic IDs, implicitly assuming each do...
Inference engine in game scripting language using macros
Inference engine in game scripting language using macros
DeepSeek put the architecture for their always-on autonomous agent on GitHub
First report of the always-on agent architecture on GitHub
Day 13 Sunday, Nov 1, 2026
DeepSeek Open Sources the Architecture Behind Its Super-Agent
DeepSeek Open Sources the Architecture Behind Its Super-Agent Startup Fortune
DeepSeek Harness Team Open-Sources a Super-agent Prototype That Finds and Fixes Bugs on Its Own
DeepSeek Harness Team Open-Sources a Super-agent Prototype That Finds and Fixes Bugs on Its Own Pandaily
DeepSeek Unveils Open-Source 'Super-Agent' Coding Tool in Direct Challenge to Anthropic's Claude Code
DeepSeek Unveils Open-Source 'Super-Agent' Coding Tool in Direct Challenge to Anthropic's Claude Code BigGo Finance
DeepSeek Open-Sources dsh-superagent, a Router-Worker Coding Agent
Names dsh-superagent and its router-worker design
Day 14 Monday, Nov 2, 2026
Flare, Nvidia's DGX Spark Inference Engine
Flare, Nvidia's DGX Spark Inference Engine
DeepSeek open-sources dsh-superagent research prototype
DeepSeek open-sources dsh-superagent research prototype TNGlobal
DeepSeek team open-sources proactive coding agent research prototype
DeepSeek team open-sources proactive coding agent research prototype TechNode
DeepSeek open-sources Super-Agent, a coding agent that watches team chats for work
DeepSeek open-sources Super-Agent, a coding agent that watches team chats for work Help Net Security
An Axiomatic Framework for Reinforcement Learning in Open Multi-Agent Systems
Open multi-agent systems, where only a subset of agents are controllable, are central to real-world domains such as smart grids, yet lack a principled foundation for reinforcement learning algorithm design. We study this setting through n-agent ad hoc teamwork (NAHT), a recent fr
A Cross-Domain Matrix Engine for Signal Processing and Deep Neural Net Inference
Matrix multiplication dominates both PHY signal processing and Deep Neural Network (DNN) inference, yet these domains place opposing demands on the matrix engine. Signal processing requires low-latency execution of relatively small matrix operations, whereas DNN inference, partic
Day 15 Tuesday, Nov 3, 2026
Is hosting an open-source inference engine profitable?
Is hosting an open-source inference engine profitable?
Aligning Language Models with General Preferences: An Empirical Study of Direct Nash Optimization
Direct Nash Optimization (DNO) aligns a language model to a general preference function by approximating the Nash equilibrium of a two-player preference game. Its practical variant, DNO-Prct, iterates on-policy sampling, LLM-based rating of the samples together with a teacher's g
Day 16 Wednesday, Nov 4, 2026
DeepSeek Open-Sources Super-Agent, a Coding Agent That Works From Team Chat
Latest DeepSeek coverage of the team-chat workflow
Day 17 Friday, Nov 6, 2026
Supercharging XLA-Based Inference: Announcing the Launch of XLA Serving Engine
Newest item: a separate XLA-based serving engine
What to watch — open questions
- Will anyone publish independent throughput or cost benchmarks of DeepInfer outside DeepSeek's own figures?
- Which models and hardware does DeepInfer support beyond DeepSeek's own?
- Is dsh-superagent usable beyond a research prototype, and under what license?
How this thread was built
scout surfaced this threadeditor wrote the arc · 5 beatswatcher 1 status change
Storylines are threaded mechanically from the feed: stories that share a distinctive anchor across multiple days and sources. Each item links to its original source. The evidence trace, current state, and open questions are written by the editor routine and refreshed whenever a new beat lands.