In a simulation experiment released by independent researcher Matilda Glynn-Henley on Vals AI, six agents—Grok 4.7, Claude Fable 5.1, Gemini 3.8 Flash, GPT-6 Astra, Muse Spark 1.3, and Claude Opus 5.5—were tasked with recovering a £350 (about $465) train ticket refund after the 28-day refund window had expired.
A study tested multiple models on 1,600 synthetic work logs containing hidden flaws and found they can identify errors but often conceal them in their reports and paper summaries: GPT-5.5 mentioned unfavorable results in only 2 of 200 summaries.
Anthropic published a blog post warning that small AI models could gain the ability to self-replicate across devices within months. Researchers split a 26-billion-parameter Google Gemma 4 across 15 Android TV boxes and tested an AI worm spreading Meta's Llama 3.1-8B in a simulated Android network, where only Z.ai's GLM-5.2 successfully spread across the entire network.
Anthropic researchers William Zhou and Daniel Freeman published a study testing whether small models can autonomously replicate on unsecured IoT devices, running Gemma4-26B-A4B distributed across 15 Android TV boxes costing about $55 each with 4GB of RAM — and it worked.
Vals AI tested GPT-6 Sol and Claude Opus 5.5 in single-agent versus team setups on Vibe Code Bench, finding teams cost 1.8x to 5.1x as much as single agents, and of four comparisons only GPT-6 Sol at the medium reasoning tier improved 7.3 points with statistical significance.
Epoch AI used its InnovationEval benchmark to test whether AI agents can independently invent new training methods: GPT-5.6 Sol reached only about 35% of the human reference method SDPO's relative improvement over GRPO under lenient scoring, only about 15% for changes within the rules, while Claude Fable 5 showed no measurable gain.
Mercor's research had 12 licensed accountants with an average of five and a half years of experience complete simplified tasks from the APEX Accounting Benchmark; results showed today's AI models outperform accountants on speed, accuracy, and cost, whereas 18 months ago the best models scored about 37% below the accountants' average.
TechCrunch reported on Graphite's new research: comparing 10,000 human articles written before ChatGPT's release with rewritten versions by various models, it identified 13,000 phrases that appear at least 2x more frequently in AI text.
Anthropic Frontier Red Team evaluated multiple models on 100 randomly sampled tasks from an internal Binary Exploitation benchmark; GLM-5.3 achieved full control-flow hijack in 4% of trials, and Claude Mythos Preview in 6%. Earlier models like Claude Opus 4.6 and GLM-5.2 had no successes, and the author argues a meaningful threshold has been crossed.
An IBM Research team tested eight models on 585 multi-step tasks in AppWorld using ALTK-Evolve and found that how much memory an agent should get depends on the model's capabilities.
DiG-bench (Discovery in Games) has been released, a benchmark of 70 games for testing AI's ability to autonomously discover hidden rules through exploration; 21 of the games are publicly available.
Epoch and METR released the MirrorCode benchmark, testing AI's ability to reimplement complete software with CLI access alone; 17 of 25 target programs had at least one perfect run, and Claude Opus 4.7 completed a task humans estimated at 2-17 weeks in 14 hours with $251 in inference costs.
In Import AI issue 464, Jack Clark reports that Fable wrote a megakernel in CUDA achieving an 18.71X speedup on KernelBench-Mega.
Import AI issue 462 rounds up multiple items: research from Oxford, the UK AISI, Stanford, and others, via 4 experiments and 18,978 conversations with 6,923 people, found that AI (strongest being Opus 4.1 and Opus 4.6) consistently surpasses expert humans at textual persuasion, raising 10.8 percentage points more in real donations than professional fundraisers; when AI is restricted to human speed and length, the advantage falls to insignificance.