Wagtail 8.0, released by the Wagtail project, introduces a new v3 API that exposes over 50 admin operations for automation, enabling AI agents and scripts to manage content without the admin UI. The A…
A developer seeking open source coding LLMs for a 128GB Ryzen AI Max system reports that Gemma 4 and Qwen 3.6 underperform, while community members recommend Qwen 3.5 35B and 122B, DeepSeek V4 Flash, …
DeepSeek V4 Flash scored 30/42 on IMO 2026 problems in Cline, clearing the 29-point gold medal cutoff for just $0.12, roughly 140 times cheaper than Claude Fable 5. The benchmark, run by Cline, tested…
CTGT founder and CEO Cyril Gorlla found that the anonymous coding model Ox Alpha, which appeared on OpenRouter on August 20 without a named developer, heavily suppresses answers about seven topics tie…
Ox Alpha, a model that appeared on OpenRouter on August 20, is behaviorally fingerprinted as part of the GLM family, with an exact 11-of-11 tokenizer match to the GLM-5.x vocabulary. The model censors…
A hardware enthusiast is weighing two multi-GPU inference builds, comparing Nvidia RTX 4000 Pro Blackwell GPUs against AMD R9700 Pro GPUs, with the Nvidia option consuming 140 watts per GPU versus 300…
DeepSeek V4 Flash, a mixture-of-experts model, ran at roughly 10 to 11 tokens per second on a single RTX 3090 with 192GB of system RAM in tests by FreeToken's desktop app, while a dense Qwen 3.8 27B m…
OneTriangle, a Y Combinator Summer 2026 company, launched hosted DeepSeek V4 Flash inference on August 24 at $0.15 per million input tokens and $0.35 per million output tokens, with CEO Hannah Chung c…
OpenCode reported that users processed 26 trillion tokens through the anonymous AI model Ox Alpha during its first four days, reaching 327,000 unique users and 8,328,244 completed sessions, making it …
As of August 2026, the best open LLM to run locally depends on VRAM, with gpt-oss-20b recommended for 8-12GB, Gemma 4 31B or Qwen3.6-35B-A3B for 24GB, and gpt-oss-120b or DeepSeek V4 Flash for 128GB u…
A new 478 KB vector file enables 'abliteration without the weights' by removing refusal directions in activation space at inference time, allowing cybersecurity defenders to run capable models on thei…
An engineer argues that enterprises are overpaying for AI by defaulting to frontier models for tasks that cheaper models handle equally well. Comparing Claude Fable 5 with DeepSeek V4 Flash, the post …
A new benchmark testing eight headless coding-agent CLIs on a single Python task found that seven of eight agents passed on both runs, but list prices per run varied 18-fold, from $0.0165 (DeepSeek V4…
A production AI tutor's context engineering experiments, detailed in the newsletter LAI #139, found that reducing tokens via summarization increased costs by roughly 2x despite sending 41% fewer token…
A single AMD MI300X GPU can serve 32 concurrent coding agents running DeepSeek V4 Flash, delivering 582 generation tokens per second after tuning, according to a benchmark by developer Ryan Zhou. The …
An anonymous AI model called Ox Alpha, listed on OpenRouter under the provider name 'stealth' on August 20, is offering free access with a 1,048,576-token context window and reported strengths in codi…
A benchmark comparing AI coding harnesses—Codex, JCode, Pi, OpenCode, and DSH—using the same prompt and the DeepSeek V4 Flash model produced completely different results on the Ship Simulator Harness …
DeepSeek V4 Flash, a 284B-parameter mixture-of-experts model with 256 routed experts and FP4 expert weights, was successfully deployed on eight AMD Radeon AI PRO R9600D GPUs (32 GB each, 256 GB total)…
In a blind comparison, three AI models preferred machine-generated literary passages over works by famous authors more than 90% of the time, with DeepSeek V4 Flash choosing machine text in 94% of deci…
The HTB-Challenger Benchmark evaluates large language models' ability to find and exploit security vulnerabilities using selected Hack The Box challenges of varying difficulty, according to the benchm…