Blog
Latest news and updates
Jev: the decision model that doesn't generate text — what it is, what it costs, and what's still unproven
A week after its September 15 launch, the model the AI developer community keeps arguing about is the one that refuses to talk. **Jev**, from San Francisco startup TypeSafe AI, generates no prose, no
DeepSeek V4.1 Flash is live: what the benchmarks show, and where it still trails V4 Pro
DeepSeek released V4.1 Flash on September 10, together with an unusual claim: the smallest model in its new architecture family performs better than its own flagship. The company says V4.1 Flash beats
DeepSeek cuts Flash prices by up to 60% — and the cache-hit line is the real story
Less than a month after raising API prices, DeepSeek is cutting them again. The company announced that from **September 10, 2026, 12:00 Beijing time**, the Flash series moves to new pricing — with a h
GPT-6 Astra: OpenAI says 'welcome to the AGI era' — here's what actually changed
On September 4, OpenAI shipped its sixth-generation flagship, **GPT-6 Astra** — and the marketing did not stay subtle. The release was framed around *"Welcome to the AGI era,"* OpenAI's Greg Brockman
DeepSeek Harness RCE vulnerability (CVSS 9.8): what's known and how to check your install
The hottest new agent framework just got its first critical security flaw. An **unauthenticated remote code execution (RCE)** vulnerability in DeepSeek Harness was disclosed this week, rated **CVSS 9.
DSH rc.8 changed its SQLite storage format — a heads-up before you upgrade
The most consequential line in DeepSeek Harness's rc.8 release isn't a feature. It's buried under "Chores": the SQLite backend got faster reads, writes, and forks, with smaller storage — **but the dat
DeepSeek Harness v0.1.1: Files API image uploads, sandbox hardening, and the vision model lands
DeepSeek Harness has moved past 0.1.0. The **v0.1.1** line landed across two release candidates this week, and it's a tight, focused pair: the vision model arrives in the adapter, image handling gets
DeepSeek makes weekends all off-peak: no peak pricing on Saturday and Sunday
DeepSeek is tweaking its peak/off-peak pricing again — and this time it's a clear win for developers. Starting **August 23, 2026, 00:00 Beijing time**, weekends no longer distinguish peak from off-pea
DeepSeek Harness turns Claude Code and Codex into sub-agents — and claims the scheduling layer
The most consequential thing DeepSeek Harness did this week wasn't a feature. It was a positioning move: Harness can now run **Claude Code** and **Codex** as sub-agents. That single capability re-fram
DeepSeek Harness v0.1.0-rc.8: 14 changes, native image input, and a weekly shipping cadence
DeepSeek Harness shipped **v0.1.0-rc.8** this week, its first major post-beta update: 14 changes, with multimodal input as the headline. But the more telling story isn't any one feature — it's the cad
DeepSeek V4-Flash-Vision-Exp: multimodal for agents, at Flash pricing
This week DeepSeek quietly dropped a new experimental model: **V4-Flash-Vision-Exp**, the first vision-capable member of the V4 lineup. It's "Exp" — experimental — but the economics are already concre
DSH v0.1.0-rc.7: the settings.plugin.item slot is now keyed — a migration note
If your plugin draws a card in **Settings → Plugins → Configurable**, v0.1.0-rc.7 breaks you. The `settings.plugin.item` slot changed from a list to a keyed slot, and the registration shape changed wi