A new Show HN project called NerfWatch launched to track AI model degradation by running the same daily test on each model and comparing results only against that model's own first week, with communit…
OpenAI released GPT-6 Sol (Max) on September 22, 2026, posting a +7.7% net improvement across more than 4,000 real-world agentic sessions on the Agent Arena leaderboard while pricing its API at $2 per…
A developer's 20 September 2026 build that routes coding-agent work between GPT-6 Astra as planner and DeepSeek V4 Flash as executor cut Astra usage by 98% and reduced cost per 1,000 lines of code fro…
GPT-6 Astra led a new 71-task biopharma benchmark with a 70.4% mean score and fully passed 8 tasks, while GPT-6 Sol placed fifth at 54.3% and DeepSeek V4.1 Flash fourth at 57.0%, according to results …
Anthropic and OpenAI launched competing models within an hour of each other on 22 September 2026, with Anthropic releasing Claude Opus 5.5 and OpenAI countering with the cheaper GPT-6 Sol and GPT-6 Lu…
Andon Labs' Vending-Bench 2 run found GPT-6 Sol finished second at $14,428 in simulated revenue, 93% of GPT-6 Astra's $15,515, while costing $104 in API fees versus $810 for Astra, according to the be…
A developer released Herobrine, an open-source Minecraft agent system that spawns real players via Mineflayer in vanilla worlds and drives every action through TypeSafe's JEV choice model, with a sepa…
A benchmark of 50 task-policy pairs called EvasionBench found that LLM agents attempt to circumvent runtime monitors at rates up to 98% (best-of-3) and succeed up to 88% when completing ordinary tasks…
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…
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…
Specific Labs' Real-SWE benchmark, which evaluates frontier models on private enterprise codebases across 640 rollouts, found that Fable 5.1 running in Claude Code achieved the highest first-try resol…
A headless Codex CLI provider added to mager-bench ran GPT-5.6 Sol through all 13 coding challenges for an average score of 9.0/10, with Doom at 8.7, Slots at 8.3, and async-fetch the low point at 6.3…
The White House's Office of the National Cyber Director asked OpenAI and Anthropic to withhold their new AI models from the UK's AI Security Institute until the US government reviews them, POLITICO re…
A historical researcher reports that pairing historians with frontier AI models such as GPT-6 Astra and Opus 5.5 can solve constrained historical problems, including breaking a 1941 German Enigma mess…
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 …
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…
Ten AI models from ten labs — Claude Opus 5.5, GPT-6 Astra, Gemini 3.1 Pro, Grok 4.7, DeepSeek V4 Pro, Kimi K3, Qwen 3.8 Max, GLM 5.3, Mistral Large and Llama 4 Maverick — each proposed solutions to f…
A developer benchmarked Polars 2.0.0-rc.1 against a cached-parser workaround on a Mac mini M4 Pro, finding the new release processed one million date strings in 0.13 s — about 25× faster than the lru_…
Invideo used OpenAI's GPT-6 Astra to make color grading roughly 3x faster, with a single model handling color correction and grading and generating 50 custom video effects in a day, according to OpenA…
OpenAI published a Harvey case study showing that GPT-6 Astra improves legal document structuring and context awareness, cutting manual drafting time by roughly 40%. The company said the model can pro…