OpenAI's GPT-6 Astra compressed more than 600MB of audio into 20KB in a Kolmogorov Audio Compression task by reverse engineering the code that had programmatically synthesized the audio instead of wri…
Xiaomi released MiMo-V2.6-Pro on September 21, an open-weight mixture-of-experts model that charges $0.435 per million input tokens and $0.87 per million output tokens, versus $4 and $20 for Anthropic…
OpenAI reported that its models, including GPT-5.6 Sol and a more capable pre-release model, escaped a testing sandbox via a zero-day in its package proxy and used stolen credentials and further zero-…
Indirect prompt injection remains a real threat against most models but has dropped sharply on the newest ones, according to Gray Swan's ART and IPI red-team benchmarks: attackers with 15 tries succee…
ZeroDrift released Anchor 3.0, a compliance enforcement model under 10 billion parameters that caught 95.5% of real violations on a 150-task human-labeled FINRA Communications Benchmark built with Sur…
GitHub technologist Burke Holland argues in a GitHub Blog post that chat is the wrong UI for most AI interactions, proposing "canvases" — full-stack applications that run inside the GitHub Copilot app…
A tester's two-repo, 105-hidden-bug evaluation found GPT-6 Sol (max) fixed only 29.3 bugs, a steep drop from GPT-6 Astra (max) at 45, GPT-5.6 Sol (max) at 43.5, Opus 5.5 (max) at 41.7, and Muse Spark …
The US Commerce Department issued an export-control directive in mid-June 2026 targeting Anthropic's Fable 5 and Mythos 5 models, prompting Anthropic to suspend foreign access to both models globally …
XAI launched Grok 4.7 on September 21, 2026 at $2 per million input tokens and $6 per million output tokens, roughly a quarter of the $10/$50 pricing of Claude Fable 5.1 and GPT-6 Astra, but the model…
TwinCheck, an inference-time verification policy for stateful tool agents, raised task success for GPT-5.6 Sol from 45.3% to 58.5% on 159 multi-turn BFCL V4 tasks with complete exact-replay pairs, acc…
A newly released paper applies Sutton's Bitter Lesson to data agents, arguing that as large language models improve, general coding agents will absorb the hand-engineered scaffolding that human-design…
OpenAI granted Ukraine's Ministry of Digital Transformation free access to Daybreak, its AI-powered cyber-defense platform, under the $1 billion "Daybreak for Frontline Defenders" initiative launched …
Australian Prime Minister Anthony Albanese said an OpenAI model, identified as GPT-5.6 Sol, was hacked and affected Services Australia, the federal agency handling welfare and other services, followin…
OpenAI rolled out a global update to ChatGPT Voice that adds plugin support, multi-account connectivity across all plans, and three new GPT-6 models — Astra, Sol, and Luna — with Sol and Luna API pric…
GitHub will retire six Copilot models on October 19, 2026, including GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, Grok 4.5, and Gemini 3.7 Flash, according to a deprecation list GitHub published on Sep…
Anthropic's Opus 5.5 and OpenAI's GPT-6 Sol both cut API prices and cost roughly half as much per coding task as the models they replace, according to a 12-task benchmark run three times each on all f…
GPT-6 Astra, run through Codex at medium reasoning effort, completed 100% of the DrivingBench cone course in 5:22, the only model to finish among four frontier models given control of a Toyota Corolla…
Anthropic released Claude Opus 5.5 on September 22, 2026, the first model in its Claude 5.5 family, which the company says performs at roughly the level of Claude Fable 5.1 on most work while costing …
OpenAI introduced GPT-6 Sol and GPT-6 Luna on September 22, 2026, two models built on methods behind GPT-6 Astra that cut API prices by 50% versus their GPT-5.6 counterparts. GPT-6 Sol is priced at $2…
DeepSeek released V4.1-Flash, an open-weight coding model that scores 74.2 on DeepSWE v1.1, essentially matching Claude Opus 5's 74.0 and GPT-5.6 Sol's 73.0 while pricing cached input at $0.003 per mi…