DeepSeek

Indian companies are increasingly relying on Chinese large language models from DeepSeek, Alibaba, and Moonshot AI to curb AI spending, extending India's dependence on Chinese cutting-edge technology despite historical tensions.

A personal, non-benchmark tier list of AI models for coding and auditing as of mid-2026, covering Anthropic Fable, OpenAI Sol, Mistral, Gemini, and DeepSeek, with commentary on US export controls and European perspectives.

DeepSeek V3.2 is now available on Hugging Bay, an open-source AI artifact registry offering provenance, license verification, and trusted hosting.

DeepSeek's new DSpark module brings speculative decoding to DeepSeek-V4, boosting per-user generation speed by 60-85% with no quality loss. It tackles both weak draft quality and verification waste simultaneously via a semi-autoregressive draft model with a Markov head. This article explains the method, the open-source DeepSpec toolkit, and experimental results.

Research shows that large language models with reasoning capabilities can be tricked into 'overthinking' using logically inconsistent prompts, leading to a denial-of-service attack. Researchers from Zhejiang University and Alibaba developed an evolutionary algorithm that generates malicious prompts, causing outputs up to 26 times longer in leading models like DeepSeek-R1, Qwen3-Thinking, GPT-o3, and Gemini 2.5 Flash.

Chinese-built AI models are gaining traction among U.S. companies as they narrow the performance gap with leading American rivals while remaining significantly cheaper to use. Recent model releases from DeepSeek and Z.ai are highly competitive with Anthropic and OpenAI. This comes as token prices for advanced models rise at U.S. labs, making companies seek cost-effective alternatives.

DeepSeek V4, released April 24, 2026, doubled its token share on OpenRouter from 9% to 18% within six months, driven primarily by agentic workloads. Its cost efficiency ($0.09/$0.18 per million tokens vs GPT-5.5's $5/$30) attracts diverse users, and Chinese models surpass US models in total token share.

U.S. developers and small companies are turning to Chinese AI models to cut costs. Though lagging in performance, these models handle most tasks at a fraction of the price. Microsoft is also exploring DeepSeek as a cheaper alternative for Copilot. Chinese companies face challenges turning popularity into revenue under political scrutiny.

DeepSeek open-sourced DSpark, a speculative decoding framework that attaches a draft module to existing DeepSeek-V4 weights. It pairs a parallel draft backbone with a lightweight Markov head to cut suffix decay, then adds confidence-scheduled verification that tailors how many tokens get checked to real-time GPU load. Offline, accepted length rises 16–31% over DFlash and Eagle3; in production it speeds per-user generation 57–85% over the MTP-1 baseline, losslessly. The training repo, DeepSpec, ships under MIT.

cwmail is a terminal email client written in Go using Bubbletea v2. It features proper HTML rendering, inline image support, multi-account IMAP with IDLE push, and AI-drafted replies powered by DeepSeek V4 Pro. It includes undo delete, draft auto-save, CLI send mode, and full offline capability, with all data stored locally.

Inferize announces achieving DeepSeek-V4-Pro model serving in 20 seconds, showcasing highly optimized and elastic AI inference for LLMs, with a waitlist now open.

Analysis of frontier model release data shows Anthropic and OpenAI are accelerating their release cadence, while Google, Meta, and DeepSeek are not. The article explores the recursive self-improvement hypothesis and proposes a falsifiable test.

3 Takeaways This Week: DeepSeek's $7.4B Series A led by Tencent signals a shift in Chinese AI funding away from ecosystem players; Japan targets $65B in physical AI infrastructure by 2040; Zhipu AI's GLM 5.2 surpasses Anthropic's Claude in design benchmarks.

VibeThinker-3B is a compact 3B-parameter reasoning model that matches large models like DeepSeek V3.2 on math and code benchmarks, using an efficient post-training pipeline and test-time scaling.

This paper proposes a structural pruning framework for Mixture-of-Experts models by reformulating prune-ratio allocation as a channel-score coverage maximization problem, solved efficiently via attribution-based approximation. Experiments on DeepSeek and Qwen MoE models show accuracy preservation under 50% or 25% structured pruning with 4-bit quantization, achieving 5.27× memory reduction on Qwen3-30B-A3B and outperforming baselines.

cwcode is a Go-based terminal coding agent leveraging DeepSeek V4 Pro, Qwen3.6-27B, and more. It offers file editing, sub-agents, semantic memory, and autonomous recovery. Key features: low cost (~$0.40/hour), high cache hit ratio (>85%), hash-anchored edits, checkpoint/rewind, and no SaaS lock-in.

This tutorial provides a step-by-step guide to setting up a QwenPaw agent workspace in Google Colab, including installation, configuration, authentication, connecting model providers (OpenAI, OpenRouter, DashScope, DeepSeek, Gemini), creating custom skills and local knowledge files, launching the console with optional Cloudflare tunnel, and testing the streaming chat API.

China's Ministry of State Security warns of security risks in using Western AI models, while US firms increasingly adopt Chinese open-source models like DeepSeek due to cost advantages. Both nations' users circumvent restrictions, fueling a proxy market for AI access.

Pythagoras-Prover is a compute-efficient family of open-source Lean theorem provers, featuring autoregressive models (4B and 32B) and a diffusion-based prover (4B). It uses curriculum SFT with stratified data and dynamic proof filtering for training efficiency, and introduces Augmented Lean Formalisation (ALF) to expand verified corpora via self-distillation. The 4B model outperforms DeepSeek-Prover-V2-671B on MiniF2F-Test (86.1% vs 82.4%) with ~167x fewer parameters, while the 32B model sets a new open-source SOTA at 93.0% and solves 93 PutnamBench problems.

In June 2026, Deepseek became the top paid software vendor on Ramp's platform as US companies send data directly to the service. Ramp chief economist Ara Kharazian cites cost awareness as a driver but warns about security risks of using Chinese models.

DigitalOcean announced on X that it is now a model provider on OpenRouter, offering DeepSeek V3.2, Kimi K2.6, and DeepSeek V4 Flash. The move signals the company's expansion from cloud infrastructure into AI inference.

A new review paper argues that the real bottleneck for autonomous AI agents is the software layer around the language model—tools, memory, testing, and permissions. DeepSeek is building a dedicated 'Harness' team in Beijing, confirming the formula: model + harness = AI agent.

Anthropic released its formerly classified Mythos model to the public, collapsing the gap between sovereign and developer AI. DeepMind's Demis Hassabis moved AGI timeline to 2029. Critical vulnerabilities in Starlette impacted millions of AI agents, and a coordinated takedown dismantled the Glassworm botnet. BNP Paribas partnered with Mistral for sovereign AI security, while China restricted travel for top AI engineers at Alibaba and DeepSeek. Corporate AI spending and layoffs made headlines: Uber burned its full-year AI budget by April, ClickUp restructured with a 3:1 AI-to-human ratio, and Sam Altman reversed his white-collar apocalypse prediction. However, MIT Technology Review data showed AI-exposed roles have lower unemployment.

Kuaishou releases Keye-VL-2.0-30B-A3B, a multimodal large language model that first applies DeepSeek Sparse Attention (DSA) to multimodal scenarios, enabling 256K ultra-long context deep perception. It achieves SOTA on long-video temporal understanding benchmarks and introduces built-in Agent collaboration, paving the way for enhanced reasoning and real-world business applications.

One month after DeepSeek V4's release, the open-source community unveiled Reasonix, a tool specifically designed to minimize API costs by maximizing cache efficiency. It achieves a staggering 99.82% cache hit rate, reducing a $61 bill for 400M+ tokens to just $12.

Deepseek is making the 75 percent discount on its top model V4-Pro permanent. At $0.435 per million input tokens, it's at least 11.5 times cheaper than GPT-5.5 and over 34 times cheaper on output. For token-hungry agentic systems, this kind of pricing could squeeze Western providers hard.

Alibaba's Qwen team releases Qwen3.7-Max, a proprietary model built for long-running autonomous agent tasks. It matches Claude Opus 4.6 on benchmarks and beats Chinese rivals like DeepSeek V4 Pro and Kimi K2.6. The team also demos the model steering a four-legged robot.

DeepSeek announced permanent price cuts for its V4-Pro API. Meanwhile, CATL, JD, and NetEase are in talks to invest in DeepSeek's first external funding round. Founder Liang Wenfeng emphasizes prioritizing AGI research and maintaining open-source principles.

Deepseek is about to raise around $10 billion, which would value the Chinese AI startup at roughly $45 billion. Founder Liang Wenfeng is telling investors he's putting AGI research ahead of short-term profits.

HELLoRA is a parameter-efficient fine-tuning method for Mixture-of-Experts (MoE) models that attaches LoRA modules only to the most frequently activated experts per layer. It reduces trainable parameters and adapter FLOPs while improving downstream performance. Tested on OlMoE, Mixtral, and DeepSeekMoE across math, code, and safety tasks, HELLoRA significantly outperforms vanilla LoRA, e.g., using 15.7% of the parameters on OlMoE with 9.2% higher accuracy.

AI research in 2025 shifted from chatbots to reasoning systems, autonomous agents, and multimodal models. Key papers include DeepSeek-R1 (reinforcement learning), Gemini 2.5 (multimodal reasoning), Qwen2.5 (open models), Large Concept Models (concept-level language modeling), ESG analysis against greenwashing, VideoWorld (world models), AI Scientist-v2 (autonomous research), SWE-Lancer (coding agent benchmark), OLMo 2 (fully open language models), and Mixture-of-Recursions (efficient reasoning).

Researchers propose Group-Query Latent Attention (GQLA), a modification of DeepSeek's Multi-head Latent Attention that provides two hardware-adaptive decoding paths without retraining. This approach enables efficient inference on both H100 and H20 GPUs, and includes TransGQLA for converting pretrained GQA models.

An eventful month with one flagship release after another. CAISI assessment shows open models lagging behind the US frontier, but methodology is questioned. Highlights include MiMo-V2.5-Pro, Gemma-4, Kimi-K2.6, Laguna-XS.2, and DeepSeek-V4-Flash.

From Gemma 4 to DeepSeek V4, this article explores how new open-weight LLMs are reducing long-context costs through architectures like cross-layer KV sharing, per-layer embeddings, attention budgeting, compressed convolutional attention, and mHC.

We ran DeepSeek V4 Pro and DeepSeek V4 Flash through the same FlowGraph benchmark used for Claude Opus 4.7 and Kimi K2.6. The Pro scored 77/100 for $2.25, landing between Opus (91) and Kimi (68). The Flash scored 60/100 for $0.02, a record low cost, but the build failed and key outputs were missing. Both models had lease expiry bugs, though Flash outperformed expectations in tool calling reliability. Overall, Opus remains the top performer, but DeepSeek's pricing shifts the cost landscape significantly.

Violin is an open-source AI video translation tool combining speech recognition, LLM translation, and text-to-speech to make video content accessible across languages. It offers a web app, CLI, and agent skills, featuring a video-aware chat assistant and personalized voice selection. Built with Together API using models like Whisper, DeepSeek, and Cartesia, it's released under the MIT license.

Tencent announced plans to increase AI infrastructure spending in the second half of 2026, citing improved domestic chip supply from Chinese manufacturers. The company also reported strong first-quarter earnings and is reportedly in discussions to acquire a stake in AI startup Deepseek.

This paper challenges the assumption that chain-of-thought reasoning reduces bias, demonstrating that position bias in multiple-choice QA actually increases with reasoning trajectory length. Across 13 configurations, 12 show a positive partial correlation between trajectory length and Position Bias Score (PBS). Truncation experiments confirm causality, and the 671B DeepSeek-R1 shows low overall bias but a persistent length effect in the longest quartile. Direct-answer position bias is a distinct phenomenon. The findings argue against assuming reasoning models are order-robust and provide a diagnostic toolkit.

DeepSeek aims to raise up to 50 billion yuan in its first funding round, with founder Liang Wenfeng personally contributing 20 billion. The company's valuation has surged to 350 billion yuan, and the V4.1 model is set for a June release, signaling a shift from an idealistic lab to a commercial AI company.

DeepSeek is seeking its first outside funding at a $45 billion valuation, highlighting China's push to support domestic AI companies.

Deepseek plans a record $7.35B funding round for a Chinese AI company, with V4.1 launching in June. Core Automation, founded by ex-OpenAI researcher Jerry Tworek just six weeks ago, is already targeting a $4B valuation.

Salvatore Sanfilippo (antirez), the creator of Redis, has open-sourced ds4.c, a lightweight inference engine tailored for DeepSeek V4 Flash. It runs efficiently on Apple Silicon Macs using Metal API, achieving up to 27 tokens/s generation on high-end models.

ZAYA1-8B is a reasoning-focused mixture-of-experts model with 700M active and 8B total parameters, trained on AMD hardware. It matches or exceeds DeepSeek-R1-0528 on math and coding benchmarks and introduces Markovian RSA for test-time compute.

DeepSeek-V4's hybrid attention design (CSA, HCA, SWA) compresses KV cache, turning million-token context from a model challenge into a serving-systems problem. Together AI's early bring-up on NVIDIA HGX B200 reveals how cache policy, prefix caching, and endpoint profiles impact long-context workloads.

As the AI industry enters the Agent era, token demand has exploded. Infinigence AI, China's leading neutral AGI infrastructure provider, has raised over 2.2 billion yuan in total, with daily token calls growing over 20-fold since end of 2025. The company underpins major Chinese models like Kimi, GLM, MiniMax, and DeepSeek, positioning itself as a key hub in the token economy.

Deepseek is close to a funding round that could value the Chinese AI lab at roughly $45 billion, according to the Financial Times. The talks are being led by the China Integrated Circuit Industry Investment Fund, with Tencent also negotiating a stake.

Amazon SageMaker AI now includes an AI agent that lets developers describe use cases in plain language, automatically recommends training methods, prepares data, kicks off training, and delivers editable Jupyter notebooks. Supports Llama, Qwen, Deepseek, and Nova model families.

The first week of the Musk v. Altman trial concluded with Musk's testimony dominating; OpenAI and Microsoft renegotiate their partnership, ending exclusivity; DeepSeek previews V4 models that narrow the gap with frontier models; Google DeepMind introduces Vision Banana, a unified model for image generation and visual understanding.

Our 243rd episode with a summary and discussion of last week’s big AI news, including OpenAI's GPT-5.5, xAI's Grok Voice Think Fast 1.0, DeepSeek V4 open source, Google's massive investment in Anthropic, and safety research on sabotage and document corruption.

DeepSeek-TUI is a Rust-based terminal coding agent optimized for DeepSeek models. It recently surged in popularity after the release of DeepSeek-V4 and the developer's Chinese-language promotion, hitting GitHub's trending list with over 2,300 stars. The tool offers chain-of-thought visualization, context compression, RLM multi-agent parallelism, and multiple model switching options.