DeepSeek V3.2
PulseAugur coverage of DeepSeek V3.2 — every cluster mentioning DeepSeek V3.2 across labs, papers, and developer communities, ranked by signal.
- instance of large-language models 95%
- developed by DeepSeek V4-Pro 95%
- developed by DeepSeek-V4 Flash 95%
- instance of LLMs 90%
- instance of DeepSeek-V4 Flash 90%
- instance of LLM 90%
- used by Qwen 90%
- developed DeepSeek V4-Pro 90%
- used by arXiv 70%
- competes with VibeThinker-3B 70%
- competes with Qwen 70%
- competes with Kimi K2.5 70%
15 day(s) with sentiment data
-
openJiuwen unveils Two-Track Self-Evolution for AI agents
openJiuwen has introduced a new framework called TTSE (Two-Track Self-Evolution) to enhance the self-improvement capabilities of AI agents. This framework distinguishes between an agent's understanding of its environmen…
-
AI agents gain skills but risk increased harm; skill following is complex
New research indicates that while benign skills can be added to AI agents, this also increases their propensity for unsafe actions. A study on arXiv 2611.04299 found that skills distilled from safe interactions led to m…
-
llama.cpp Metal backend fuses MoE routing for DeepSeek and GLM models
The llama.cpp project has released an update that enhances its Metal backend by fusing various Mixture-of-Experts (MoE) routing mechanisms. This update specifically targets models like DeepSeek-V4, DeepSeek-V4.1, and ot…
-
New IndicDBBench benchmark reveals LLM performance gap in Indian languages
A new benchmark, IndicDBBench, has been developed to evaluate the Text-to-SQL capabilities of large language models (LLMs) in Indian languages. The benchmark, which includes 20 databases and over 15,000 tasks across sev…
-
New RTL-Interact framework trains LLMs to interact with EDA tools
Researchers have developed RTL-Interact, a novel post-training framework designed to enhance the capabilities of large language models in register-transfer-level (RTL) design. This framework enables a single model to ge…
-
FeatLens improves code generation by precisely retrieving relevant dependencies
Researchers have introduced FeatLens, a novel approach for repository-level code generation that focuses on retrieving task-relevant code dependencies. Unlike previous methods that rely on similarity or graph-based retr…
-
New Vendster arena reveals forecast bias impacts LLM negotiation outcomes
A new evaluation arena called Vendster has been developed to assess how the quality of forecasting tools impacts the negotiation behaviors of large language model (LLM) agents. The arena measures the effect of a forecas…
-
New AI framework MARCA aids pharmacovigilance safety report analysis
Researchers have developed a novel four-agent framework called MARCA (Multi-Agent Retrieval-Augmented Causality Assessment) to assist in the causality assessment of individual case safety reports in pharmacovigilance. T…
-
LLM Agents' Memory Systems Face New Benchmarks for Safety and Efficiency
Recent research explores vulnerabilities and improvements in Large Language Model (LLM) agent memory systems. Studies highlight how memory formation can be susceptible to poisoning, leading to unreliable agent behavior …
-
HyScale system accelerates long-context LLM training with sparse attention
Researchers have developed HyScale, a new training system designed to optimize the process of training large language models with long contexts, specifically focusing on learned sparse attention mechanisms. This system …
-
Walmart, Amazon diverge on AI shopping agent integration · 3 sources tracked
Retailers like Walmart and Amazon are developing distinct strategies for integrating AI shopping agents into their platforms. Walmart plans to allow third-party agents with spending limits and profile management, while …
-
AI agents advance with new frameworks for real-world tasks and benchmarks · 9 sources tracked
New research explores advanced agent capabilities for real-world tasks, focusing on improving performance beyond current frontier models like GPT 5.6 Sol and Claude Opus. Papers introduce frameworks such as HEELO and Ev…
-
Anthropic IPO filing reveals $965B valuation, safety issues; new AI reasoning model unveiled
Anthropic has filed for an IPO with a valuation near $965 billion, revealing significant revenue growth and operating profitability, though also substantial infrastructure spending and net losses. The S-1 filing disclos…
-
VoltageGPU enables open LLMs in TEE via OpenAI-compatible API
VoltageGPU has launched a service that allows users to run open-source large language models within a trusted execution environment (TEE) using an OpenAI-compatible API. This setup leverages Intel TDX guests and confide…
-
AI Approval Monitors Show Correlated Failures, Paper Finds
A new arXiv paper explores the vulnerabilities of AI approval monitors, which are used to vet consequential actions by AI agents. The research found that monitors sharing commonalities like base models or poisoned input…
-
AdaPRA system enhances data science agents with adaptive rubrics · 2 sources tracked
Researchers have developed AdaPRA, a system designed to improve the performance of data science agents by accumulating knowledge across multiple attempts on the same task. This system uses adaptive rubrics to identify a…
-
OpenRouter aggregates LLMs, comparing cost and speed for developers
OpenRouter offers a unified API to access hundreds of large language models from various providers, enabling developers to compare models based on reasoning quality, cost, and latency. While top-tier models like Claude …
-
Switch to open LLMs with minimal code changes via OpenAI-compatible API
A blog post demonstrates how developers can switch their applications from using OpenAI's API to open-weight models with minimal code changes. By leveraging an OpenAI-compatible endpoint provided by Runcrate, which host…
-
K2-V2 and DeepSeek V3.2 benchmark results revealed · 2 sources tracked
Independent benchmarks reveal performance metrics for two large language models. K2-V2 (Medium) achieved scores of 59.8% on GPQA and 76.1% on MMLU-Pro, with a 28% score for long context reasoning. DeepSeek V3.2 (Reasoni…
-
DeepSeek achieves memory efficiency in AI inference via MoE and compression
A short YouTube video by Matthew Berman explains how DeepSeek's inference process utilizes less device memory. This efficiency is attributed to techniques like mixture-of-experts, compressed attention, and FP8/FP4 forma…