53 verified stories covering DeepSeek, product updates and industry developments.
DeepSeek is reportedly set to lead a seed round of tens of millions of dollars in Winverse AI, a robotics startup founded by ex-Tencent scientist Hu Han, at a roughly $400 million valuation.
CrowdStrike says a lone attacker likely based in China used an open-source AI pentest tool and DeepSeek to breach at least nine South Korean banks.
vLLM's new optimizations push DeepSeek-V4.1-Flash throughput up to 5.3x under rate limits, mainly by shrinking the KV cache and skipping memory offload for long contexts.
Bloomberg reports DeepSeek's new funding round may near 100 billion yuan, with Tencent and CATL leading as the AI firm preps for an early 2027 listing.
DeepSeek's new paper details DSec, the sandbox system behind its agent training that launches about 3 million sandboxes a day at peak concurrency near 380,000.
New Fireworks AI benchmark data shows DeepSeek-V4.1-Flash matching GPT-6 Astra's score on agentic coding tasks while costing 15 times less per task.
DeepSeek quietly added a voice chat button to its app in a grayscale test, offering four reading voices, while the release carries no official announcement at all.
AMD shipped its DeepSeek V4.1 Flash image two days after CUDA's vLLM, and SemiAnalysis says it delivers up to 14.8x worse performance per dollar than H200, and up to 42x worse than B200 and B300.
DeepSeek launched V4.1 Flash on Sept 10 with new pricing, and starting Sept 14 all deepseek-v4-pro requests reroute to the smaller model at its rates.
DeepSeek has reportedly hired CITIC Securities to prepare a STAR Market IPO, with a new funding round targeting a valuation of RMB 500 billion.
NSA, CISA, and the FBI say six Chinese AI firms used unauthorized, industrial-scale distillation of US models from Anthropic, OpenAI, Google, and SpaceX.
DeepSeek opened a V4.1 Flash preview with native multimodal support, but the model ID expires September 10 — under 48 hours after launch.
DeepSeek retook OpenCode's daily usage lead the moment GLM-5.3-Flash started charging, ending its 41-trillion-token free run.
DeepSeek made ¥475M in revenue but lost ¥715M in early 2026, as an 82.9% API margin gets erased by free-tier subsidies and a new peak-pricing scheme.
DeepSeek's new deepseek-v4-flash-vision-exp caps every image at 384 tokens with no extra vision fee, effectively putting a ceiling on GUI agent screenshot costs.
Liquid AI's LFM2.5-DSpark draft models bring speculative decoding to three on-device models, cutting decoding latency up to 3.18x with identical outputs.
LMSYS's DeepSeek-V4-Pro benchmarks show Nvidia's China-market H20 chip closing the decode-speed gap with the flagship B300 to just 1.42x.
DeepSeek's V4 Pro API now points to the 0813 version, pairing a 1M-token context and 384K output ceiling with a clear price warning.
DeepSeek put V4-Flash 0731 into API beta, added Responses API support for Codex, and reported large agent-benchmark gains from post-training alone.
DeepSeek reportedly develops its own inference chip is reframed for global technology readers, with the key numbers, product claims and open questions kept intact.
Check Point says a DeepSeek-generated sample can be turned into browser-only ransomware by abusing Chrome’s File System Access API.
DeepSeek plans a mid-July V4 release and will double API prices during two China workday peak windows to smooth demand.
DeepSeek and Peking University open sourced DSpark, an inference framework that raises generation speed by 60% to 85% and can lift single-GPU throughput as much as 6.6 times.
Lindy swaps Claude for DeepSeek and says costs fell by 90%
DeepSeek brings peak-and-off-peak pricing to model APIs
The case shows how export controls can become trade bargaining chips even when security agencies have already made their case.
The financing structure accepts outside capital while keeping voting power away from commercial investors.
DeepSeek closed its first outside round at a valuation above $50 billion, but commercial backers get no voting power.
Enterprise workload estimates show Claude, ChatGPT, DeepSeek, Kimi and GLM separated by nearly 9x in cost, forcing OpenAI and Anthropic toward price pressure.
Ramp and OpenRouter data show DeepSeek usage accelerating as long-running coding and data agents make token price a first-order enterprise cost.
Ramp data shows DeepSeek leading SaaS vendors by relative growth among US firms, even though its terms say personal data may be stored in China.
Reuters says DeepSeek’s first major financing could value the company at 350 billion to 400 billion yuan while its founder remains the largest investor.
China is tightening controls on overseas travel for key AI personnel at private firms like Alibaba and DeepSeek, requiring government approval before leaving the country.
DeepSeek has permanently set V4-Pro API prices at 75% off, locking in $0.87 per million output tokens and challenging rivals.
DeepSeek is reportedly discussing a first funding round led by China’s national chip fund, pushing its valuation toward $45 billion and tying its next phase to domestic compute.
The US State Department sent a diplomatic cable on April 24 instructing embassies worldwide to warn host governments that Chinese AI companies are systematically "extracting and distilling" models from US labs, naming DeepSeek, Moonshot, and MiniMax.
DeepSeek released its V4 series on April 24, featuring a 1.6-trillion-parameter Pro model, but markets showed little reaction compared to last year's shock. Analysts say the surprise factor has faded as China's open-source AI output becomes routine.
DeepSeek slashed the price of its V4-Pro model by 75% and cut cache-hit pricing across its entire API to one-tenth, just 72 hours after the preview release, intensifying the AI price war.
DeepSeek released V4 Flash and V4 Pro as preview versions on April 24, with the full release delayed until the second half of 2026 due to Huawei's Ascend 950PR chips not yet being ready for mass production.
DeepSeek released V4-Pro and V4-Flash under MIT license, achieving top competitive programming scores while pricing output at $3.48/M — a fraction of GPT-5.5 and Claude Opus 4.7.
DeepSeek is seeking its first-ever external funding round, targeting at least $300 million at a valuation exceeding $10 billion, according to four anonymous informed sources. The move marks a major shift for the AI lab, which had long rejected outside investment.
DeepSeek quietly released V3.2 with a sparse attention mechanism called DSA, cutting computational complexity from O(L²) to O(Lk) and slashing API prices to $0.28/M input and $0.42/M output, while maintaining comparable benchmark performance.
OpenAI, Anthropic, and Google have joined forces to combat distillation attacks, accusing Chinese AI companies DeepSeek, Moonshot AI, and MiniMax of using 24,000 fake accounts to extract over 16 million interactions from Claude.
DeepSeek has quietly introduced two new chat modes, Instant and Expert, with the Expert mode offering higher quality for complex tasks but removing file upload support, hinting at a tiered service structure ahead of the anticipated V4 release.
DeepSeek is set to release V4, a trillion-parameter MoE model running on Huawei's Ascend 950PR chips, marking the first frontier AI model built entirely for a domestic Chinese chip architecture and a potential shift away from NVIDIA's CUDA ecosystem.
DeepSeek has released Prover-V2, a specialized model for automated theorem proving in the Lean 4 formal verification system, achieving near state-of-the-art results on the miniF2F benchmark.
A comprehensive comparison of major AI reasoning models from OpenAI, DeepSeek, Google, and Anthropic, examining their approaches, trade-offs between depth and speed, and the open-source vs. closed-source landscape.
DeepSeek V4 and Qwen 3.5/3.6 are reshaping the AI market with massive context windows, native multimodality, and aggressive pricing, pushing open-source models from a 1% to a 15% global share in one year.
DeepSeek V4 and Qwen 3.5 (also labeled Qwen 3.6-Plus) are two Chinese open-source AI models competing on cost and capability, challenging expensive closed-source giants with trillion-parameter scale, massive context windows, and rock-bottom API pricing.
DeepSeek has achieved legendary cost efficiency in training large AI models, with V3 costing about $5.5 million and R1 GPU usage at roughly $294,000, compared to U.S. competitors spending $100 million to $1 billion for similar-scale models.
DeepSeek V3.1 is the company's first hybrid model, combining the capabilities of the general-purpose V3 and the reasoning-focused R1 into a single architecture.
DeepSeek released the R1 reasoning model as open source in January 2026, achieving scores comparable to OpenAI's o1 on math, code, and logic benchmarks while its distilled 32B variant outperformed o1-mini on several tests.
DeepSeek V3 uses a Mixture-of-Experts architecture with 671B total parameters but only 37B activated per token, achieving top-tier open-source benchmark scores while costing just $5.5 million to train.