DeepSeek
Chinese AI lab famous for the V4 series. V4 Flash at $0.14/1M is the cheapest pro-quality API. Peak/valley pricing model.
Chinese AI lab famous for the V4 series. V4 Flash at $0.14/1M is the cheapest pro-quality API. Peak/valley pricing model.
Models from DeepSeek(2)
DeepSeek History Timeline(30)
DeepSeek closes roughly $15B round
CATL reportedly put in around $1.5B with Tencent following, at a pre-money valuation near 500 billion yuan against annualised revenue of about $1B.
SourceDeepSeek updates Flash line amid domestic price war
Pricing cuts alongside MiniMax M3.1 Flash formed price competition among Flash-class models.
SourceDeepSeek open-sources DeepInfer inference framework
The framework powering DeepSeek-V4.1-Flash is open sourced with NVIDIA SM100 and Huawei Ascend 950 support. Offline inference only in this release.
SourceAdded by DHS to US civilian agency ban list
Oct 19 DHS ordered all US civilian agencies to stop using DeepSeek and to block third-party services integrating it; the directive cites concerns about potential exfiltration of US government data to China, the latest systematic federal restriction after August's risk assessment
SourceV4.1 Flash 1017 build released under MIT
Oct 17 DeepSeek released V4.1-Flash-1017 weights on Hugging Face and ModelScope under MIT; API name deepseek-flash and pricing unchanged; Terminal-Bench 4.0 rose to 42.3 and DeepSWE v1.1 to 77.0
SourceDeepSeek leads Winverse AI humanoid robotics round
DeepSeek led the Winverse AI seed round on October 13 at a roughly $400M post-money valuation, with Meituan's Long-Z Investments co-investing.
SourceDeepSeek raising at least 80 billion yuan, Tencent and CATL as largest contributors
The round could approach 100 billion yuan (about $15B) and will fund 160,000 Huawei Ascend AI accelerators at an Inner Mongolia data center and prep for a potential 2027 IPO.
SourceDeepSeek annualized revenue run rate crosses $1B
The September annualized run rate doubled in a single quarter, driven by strong demand for low-cost models after the V4 Flash release.
SourceHarness desktop enters global public preview, macOS and Windows installers open-sourced
On October 2 DeepSeek released a desktop preview of Harness v0.2 with macOS and Windows installers; the GitHub repo has about 242,400 stars; positioning shifts from coding to everyday work, with new automation task plugins and a plugin management panel; MIT-licensed.
SourceDeepSeek Harness desktop preview leaks with Apple notarization
Harness desktop client for Mac and Windows leaked to the community between September 25 and 27, already Apple-notarized; offers Standard/PTC/Minimal/Create modes.
SourceOpenCode integrates V4.1 Flash as full partner
V4.1 Flash ranks first in recent usage on OpenCode platform (~13% of 2M-level observations)
Source$7.5B Series round near completion
9/24 plans to close 50 billion yuan ($7.5B) round before end of October at 500 billion yuan target valuation; annualized revenue run rate reaches $1 billion
SourceFormer Hillhouse 90s-born partner Yan Wentao appointed first CFO
9/24 appointment to oversee finance, IR, and IPO prep; CITIC Securities engaged as STAR Market advisor
SourceDeepSeek retires V4 Flash and V4 Flash Vision Exp, cuts V4.1 Flash
Around September 23 DeepSeek stopped accepting new requests on V4 Flash and V4 Flash Vision Exp; existing IDs are silently rerouted to V4.1 Flash at the new pricing introduced on September 10.
SourceDeepSeek V4 Pro traffic routes to V4.1 Flash
From 12:00 Beijing time on September 14, DeepSeek V4 Pro API requests are silently routed to V4.1 Flash and billed at the V4.1 Flash rate, letting users pick up an upgrade and a price cut without code changes
SourceDeepSeek V4.1 Flash officially released
A 552B-parameter MoE model using the new Causal-Encoder-Decoder structure, with 8B activation for input and 16B for output. KV Cache demand for HBM drops to one-fourth and SSD to one-eighth of the prior generation. Native multimodal support. V4 Flash and V4 Flash Vision Exp are retired; from September 14 at 12:00, V4 Pro requests temporarily route to V4.1 Flash.
SourceDeepSeek nears 50B yuan funding at 500B yuan valuation
DeepSeek plans to raise ~50B yuan at ~500B yuan pre-money ($74B), expected to close before end of August, targeting 2027 STAR Market IPO
SourceV4-Flash-Vision-Exp visual understanding model launches
Native image input. Can identify people, read screenshots, parse charts. Harness updated to v0.1.1-rc.1.
SourceRMB 141M strategic stake in Unitree Robotics (2.31%)
Signed MoU covering joint AGI R&D, priority robot procurement, and AI model adaptation.
SourceHarness 0.1.1-rc.2 critical vulnerability QVD-2026-57410 disclosed
Harness 0.1.1-rc.2 critical vulnerability QVD-2026-57410 disclosed
SourceDeepSeek unifies weekend pricing to off-peak rates from August 23
Optimized peak-valley billing: weekends no longer distinguish peak vs off-peak and use off-peak rates all day. The same day V4 Pro GA hit the API with significantly stronger Agent capabilities, Responses API plus Codex support, scoring 87.9 on Terminal-Bench.
SourceDeepSeek launches multimodal V4-Flash-Vision-Exp
Experimental multimodal vision-understanding V4-Flash-Vision-Exp launched on the API, priced the same as V4-Flash. Text performance matches V4-Flash and multimodal agent capability is close to Opus 4.8. Files API launched alongside. Chartography 64.3, Terminal Bench 2.1 83.9, ApexBench rose from 26.2 to 36.5.
SourceDeepSeek Harness v0.1.0-rc.8: native image requests plus Claude Code / Codex sub-agents
Model adapter supports native image requests and image+text input. Claude Code and Codex became installable Profile Bundles; Codex gets a non-interactive permission mode and multi-instance parallelism. web_search supports concurrent queries.
SourceDeepSeek V4-Pro peak/valley pricing takes effect
From 00:00 Beijing time on August 17, peak/valley pricing takes effect. During peak hours (9-12 and 14-18 Beijing time on weekdays), V4-Pro cached input price rises up to 1100%, output from 6 to 27 yuan per 1M tokens.
SourceV4 API peak/off-peak pricing live; V4-Pro cached input +1100%
At midnight on Aug 17 DeepSeek V4 fully moved to peak/off-peak pricing; V4-Pro cached input rose up to 1100% during peak hours.
SourceV4-Pro official version live (1M context + 384K output)
Went live on App, web, and API on Aug 13 with three thinking intensity tiers.
SourcePeak-valley API pricing effective 8/17
Peak-hour V4-Pro at ¥9 input / ¥27 output per million tokens; V4-Flash at ¥3 input / ¥9 output.
SourceDeepSeek Harness v0.1 crosses 100K GitHub stars in 45 hours
Harness developer preview open-sourced August 13 (MIT license) under "everything is a plugin" philosophy based on the Cordis plugin system. Reached 105K GitHub stars in 45 hours, 4,300+ third-party plugins within 3 days.
SourceDeepSeek-V4-Flash tops OpenRouter weekly with 9.39 trillion tokens
For the week of August 3-9 DeepSeek-V4-Flash-0731 led OpenRouter global call volume at 9.39 trillion tokens.
Source