#DeepSeek
318 related articles
DeepSeek V4.1 Flash 'Minor Update': A Dangerous Signal for Closed-Source Models
DeepSeek V4.1 Flash-1017 ships as a 'minor update' but jumps 11 points on Terminal-Bench, sparking HN debate. We break down 763B params, MIT license, agentic cost advantages, and the new low-cost search/fetch API.
DeepSeek V4.1 Flash 1017 Released: Top in 4 of 11 Benchmarks
DeepSeek V4.1 Flash 1017 released, claiming first place in 4 of 11 benchmarks, leading Kimi K3 and GLM 5.3 in code and Agent tasks, surpassing Cloud Opus 5 in long-horizon tasks. This article analyzes its three capability directions and practical usability.
DeepSeek New Version Hands-On: Same Price, Costs Slashed by 60%
DeepSeek v4.1 Flash 1017 cuts DeepSWE task cost ~2.5x at the same price, driven by shorter output tokens, effort-level length penalties, and token economics.
DeepSeek V4.1 Flash Update: The Parameter Truth, Benchmark Controversy, and Cost Revolution
DeepSeek V4.1 Flash 1017 released: all 11 benchmarks up, Terminal Bench 4.0 +11 points. We unpack the 763B parameter truth, benchmark scope controversy, cost gains, and local deployment barriers.
DeepSeek V4.1 Flash Open-Source Hands-On: Faster, Cheaper, and Runs on Consumer Hardware
DeepSeek V4.1 Flash-1017 goes open source: community tests show 30/30 coding task passes at about 4.5 cents each, a local inference engine 2.1–2.3x faster than VLLM, and a 552B model running on a single 4090.
DeepSeek Launches Search API, Zhipu's Open-Source Legal LLM Tops Charts
DeepSeek releases V4.1 Flash 1017 with web search and scraping APIs; Zhipu open-sources legal LLM GLM-Legal topping VALS AI dual leaderboards; Tsinghua open-sources agent OS Pilotech 2.0 cutting task costs 83%; Claude Haiku 4.5 to be retired end of month, Codex deployment data revealed.
DeepSeek v4.1 Flash Review: Challenging GPT-6.1 at 0.44 Cents
DeepSeek v4.1 Flash 1017 checkpoint tested: benchmarks near Claude Opus 5, 0.44-cent GPT-6.1 Sol-level motion graphics, 250 tokens/s, time-of-use pricing, 763B open weights for local deployment. Benchmarks, pricing, and harness controversy explained.
DeepSeek Search and Fetch API Launches: Just $0.75 per Thousand Requests
DeepSeek launches a Search and Fetch API at just $0.75 per 1,000 successful requests. This analysis covers its pay-per-success pricing logic, impact on AI app development, and key factors developers should evaluate before integrating.
Swap Models with One Line of Code: Cut AI Agent Inference Costs by 15x
A developer's real-world test: switching the AI Agent backend to DeepSeek V4.1-Flash with difficulty-based tiered routing cut inference costs ~15x, with CSAT holding steady across 4,000+ conversations. This article breaks down the cost multiplication effect, two-line-code migration, and a one-week A/B validation path.
DeepSeek V4.1 Flash "Minor Update" Hands-On: Across-the-Board Benchmark Gains Reshape the Reasoning Market on Value
DeepSeek V4.1 Flash 1017 is called a minor update, but Terminal Bench 4.0 jumped from 31% to 42% in five weeks, at ~20 cents per task and one-sixteenth the output price of GPT 5.6 Sol.
DeepSeek V4.1 Flash 1017 Hands-On: Performance Gains and Controversy Behind a Low-Key Update
DeepSeek V4.1 Flash 1017 shipped quietly as a patch: Terminal-Bench 4.0 jumps from 31.2 to 42.3, cache shrinks to 890 bytes/token, but hard reasoning still trails Claude Opus 5.
DeepSeek V4.1 Flash Tops Open-Source Model Rankings
DeepSeek V4.1 Flash (1017) reportedly tops open-weight models per Reddit. This article explains the version naming, its impact on the open-source landscape, and key deployment, fine-tuning, and migration considerations for developers.
DeepSeek V4.1-Flash Hands-On: Benchmark Comparison with Opus 5 and GLM-5.3
DeepSeek V4.1-Flash-1017 improves Terminal-Bench 4.0 by 11 points at unchanged pricing. We compare it with Opus 5, K3, and GLM-5.3 on positioning and cost-effectiveness, analyze switching timing and migration costs for terminal task scenarios, and offer practical selection advice.
DeepSeek-v4.1-Flash-1017 Leaked: The Lightweight Model Evolves Again
DeepSeek-v4.1-Flash-1017 appears on Hacker News. This article explains its naming rules, Flash series positioning, and iteration cadence, and analyzes the practical impact on developers' model selection and migration decisions.
DeepSeek v4.1 Flash 1017: An In-Depth Analysis
The DeepSeek v4.1 Flash 1017 release has sparked debate in the tech community. Drawing on Hacker News discussion, this article analyzes its naming logic, technical positioning, and application prospects to help developers choose the right model.
China's AI Rise: DeepSeek, Robots, and the New AI Race Landscape
In-depth analysis of China's AI rise: DeepSeek's efficient models challenge compute assumptions, open weights become a strategic advantage, Huawei breaks chip limits, humanoid robots and autonomous driving accelerate, and how India should respond to this AI race.
DeepSeek Harness Step-by-Step Tutorial: Four Modes and Plugin in Practice
A step-by-step DeepSeek Harness tutorial: workspace, the four modes (PTC/standard/minimal/creation), three access permission levels, the 100+ plugin system, and turning it into an auto-launching desktop app.
DeepSeek API Outage Lasts 4 Hours 52 Minutes: The Disaster Recovery Lesson Behind a Week's Consumption Compensation
DeepSeek API suffered a partial outage of 4 hours 52 minutes on October 15. The company fully refunded 7-day consumption as bonus credits. This article outlines the outage scope, compensation details, and provides a complete code solution for configuring a backup provider for DeepSeek calls, plus an action list for this week.
vLLM Releases v0.32.0rc2: DeepSeek-V4.1-Flash Evaluation Pinned to FP8-Engram Revision
vLLM releases v0.32.0rc2, with a CI change pinning DeepSeek-V4.1-Flash-NVFP4 evaluation to the FP8-Engram revision, reflecting the inference framework's focus on reproducible multi-precision evaluation.
DeepSeek Funding Doubles to 100 Billion Yuan, ByteDance Seed Establishes LLM Safety Department
DeepSeek's second funding round doubles to 100 billion yuan; ByteDance Seed becomes China's first to establish an LLM safety and alignment department. Huawei's Kirin 2027 chip enters testing, Tesla Optimus targets year-end mass production, and Yin Zhiyao warns of AI bubble and semiconductor industry consolidation.