Fetching from the wire…
01
02
03
04
05
06
07
08
Models2026-09-17 · source-backed
Enclave published results September 16 in which the model gained code execution on all 11 targets including Grafana, Jenkins and Nextcloud, while all four correctly patched targets held. It found the six planned attack routes plus five unexpected ones. Total spend was $5.14 including failed attempts, achieved because 266.2 million of 268.3 million input tokens were cache reads. That ratio is the practical consequence of the KV cache compression DeepSeek released with V4.1-Flash on September 10, one quarter the HBM and one eighth the SSD of the prior generation. Long autonomous security work just got cheap enough that cost stops being a deterrent for either side.
Each link below shares sources, entities, or timing with this story.
DeepSeek posted a community notice: once V4.1 Flash launches around September 10 Beijing time, and until a V4.1 Pro exists, every V4 Pro request routes to V4.1 Flash and bills at Flash unit pricing. The stated reason is that Flash has surpassed Pro on performance, cost, speed...
The API changelog carries it verbatim: "In response to user demand, we have decided to continue providing API services for DeepSeek V4 Pro after September 14, 2026, with the billing method remaining unchanged." Every deepseek-v4-pro request was scheduled to be answered by V4.1...
Latent Space's September 12 roundup collects reaction to the causal encoder-decoder release: 763B total parameters, 8B active at prefill and 16B at decode, 1M context, KV cache cut to about 890 bytes per token, roughly one eighth of V4 Pro. Sebastian Raschka argues the archite...
DeepSeek-V4-Flash-0731 landed July 31 under MIT with a DSpark speculative-decoding module attached. Terminal Bench 2.1: 82.7. Toolathlon-Verified: 70.3. DSBench-FullStack: 68.7. DeepSWE: 54.4. NL2Repo: 54.2. The model card claims it beats DeepSeek-V4-Pro (Preview) "despite its...
DeepSeek dropped V4 in mid-June as an open-weight model with a 1-million-token context window, priced at $1.74 per million input tokens, posting near-parity with GPT-5.4 on math and Q&A benchmarks (MindStudio). That's the headline number. The architecture underneath is more in...
$3,054 against $38,370. Same benchmark, better score. Praxist (arXiv 2608.25955, submitted August 26) replaces per-attempt agent memory with a typed evidence graph of findings, plus lane-structured frontiers and agendas, so later attempts inherit validated mechanisms rather th...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.