Latest articles
Can AI check AI-written code? What GitHub ReviewBench measures
GitHub’s ReviewBench measures AI code review through coverage and finding quality. We explain its 219 public PRs, what 96.6% agreement means, and how to interpret the results for real development.
Technology & ResearchStrata runs a large MoE model by sharing computation between the CPU and GPU
Strata v0.1.39 speeds up local generation and adds concurrent requests and Codex CLI support. We explain the memory layout behind a large MoE on a 12 GB GPU, RAM and SSD requirements, long-prompt and concurrency trade-offs, and the limits of its Responses API.
Technology & ResearchStrands Decider 2B is a small AI model for making choices locally
A large LLM plans the work while a small model handles recurring choices. We explain how Strands Decider 2B works, how it can route tasks to local specialists, what confidence means, and the conditions behind its hardware support and performance figures.
Technology & ResearchGLM-5.3 changes attack economics and cyber defense priorities
When AI takes on exploit development and preparation, the same number of attackers may be able to try more targets. GLM-5.3 evaluations, observed AI misuse, and five Japanese incidents help explain the changing economics of attacks and the defensive work that needs urgency.
Technology & ResearchTensorFold speeds up local AI generation by drafting and verifying tokens
TensorFold runs models on Macs and supported NVIDIA GPUs. We explain draft verification, quantization, memory, and model support. Project measurements show faster single-request generation but lower throughput than vLLM at eight concurrent requests, so the workload changes the verdict.
Technology & ResearchDeepSeek releases software for AI computation and communication on Huawei chips
Running AI on chips other than NVIDIA's requires adapting computation, memory layouts, and communication between chips. The releases of DeepGEMM-Ascend and DeepEP-Ascend explain why, and clarify the scope of API compatibility. The performance figures are the developer's measurements, and the communication benchmarks used a PoC HDK that has not been publicly distributed.
Chips & InfrastructureGemini 4 Argon takes on long-running work with up to 1 million output tokens
Google has announced Gemini 4 Argon. We examine its output limit of 1 million tokens, strengths in work-related evaluations, and introductory and regular pricing. We distinguish early access for specialists from plans for general availability, and check the conditions behind Google's measurements and external evaluations.
Products & ServicesOpenAI introduces dots, an AI assistant for everyday work
OpenAI's dots handles research, scheduling, and document updates. We explore creative and analytical uses suggested by 12 apps and the computing resources observed in one cloud environment, alongside pricing, Gmail and calendar permissions, data retention and training, and the distinction between AI reporting and disclosure by its operator.
Products & ServicesA notice of 62,500 Codex credits: who qualifies and what $2,500 means
A Pro 200 subscriber received a notice of 62,500 credits. We separate that individual notice from public documentation to explain API billing, the December 31 expiry, and changes to the regular usage allowance.
Products & ServicesGPT-6.1 Sol launches into a cost contest with Opus 5.5
At half Opus 5.5's standard API rates, GPT-6.1 Sol is a contender for coding and computer use. But the price gap narrows on long inputs, and benchmarks use different conditions. We examine both companies' documentation to assess the trade-offs.
Products & ServicesWhen AI remembers you, who can read its memory?
What do device-derived keys and isolated execution protect in Google's new server-side memory design? We examine the encryption, two unresolved audit findings, and what is still unknown about availability, using primary sources.
Technology & ResearchGemini 3.8 Flash TTS arrives: A comparison with GPT-Live
Compare Gemini 3.8 Flash TTS and GPT-Live with response speed as the priority. We examine first audio versus substantive answers, language support, speech evaluations across seven languages, API pricing, differing measurement conditions, and Google's 2027 price changes.
Products & ServicesCan efficiency gains offset Codex Pro's reduced usage allowance?
The new $200 Pro multiplier is 10x Plus. We examine actual usable capacity and efficiency gains in light of Tibo's additional explanation: a higher usage baseline, temporary retention of 20x for existing subscribers, and extra credits.
ColumnsClaude Sonnet 5.5 launches with more than 30% faster output
Claude Sonnet 5.5 speeds up output while keeping API rates unchanged. We compare performance and costs in evaluations from Artificial Analysis, Vals AI, Mercor, and others, and explain availability and API migration considerations.
Products & ServicesMeta launches Muse, a personal AI agent
Muse carries out research and operates services in a personal cloud environment, requesting approval when needed. We examine its rollout, the work it can handle, and how it manages data.
Products & ServicesAnthropic says Claude discovered a previously unknown enzyme system
Researchers followed up on patterns identified by AI with further analysis and experiments. We examine what is new, what has been verified, and why ten attempts to repeat the search failed to rediscover it.
Technology & ResearchClaude Opus 5.5 launches with roughly 60% lower cost per evaluation task
Claude Opus 5.5 improves its score in an independent evaluation while reducing cost per task by about 60%. We explain performance, pricing and practical caveats.
Products & ServicesGPT-6 Sol and Luna launch with API rates cut by half or more
In independent evaluation, cost per task falls about 47% for Sol and 61% for Luna. Overall performance is near the previous generation, with gains in avoiding wrong answers and business automation but weaker results in document creation and other work.
Products & ServicesGemini 3.8 Live launches, keeping conversation going while work runs
The two models differ in handling complex requests, acknowledgments and interruptions. We explain Japanese support, the cost of a whole conversation and availability across apps.
Products & ServicesApple releases iOS 27 and the English beta of the new Siri
Using models built in collaboration with Google's Gemini, the new Siri finds information in messages and email and operates apps. We explain supported iPhones and the planned timing for Japanese.
Products & ServicesAnthropic's CEO calls for a slower pace of AI development
Leaders of companies driving the AI race endorse a proposal to slow down. We examine safety concerns, corporate interests in regulation and competition, and what would turn the appeal into an actual slowdown.
Society & PolicyGPT-6 Astra launches with improved computer-use performance
GPT-6 Astra scores higher on computer-use evaluations and completes tasks faster. We explain performance and pricing changes, availability and usage restrictions.
Products & ServicesNVIDIA agrees to acquire Hugging Face in a deal worth about $12.9 billion
Hugging Face, which supports publishing and using AI models, is set to join the chipmaker. Support for other chips and clouds is to continue. We explain the deal's components and implications for developers.
Business & MarketsAnthropic announces Claude Fable 5.1 and Mythos 5.1
Fable 5.1 outperforms its predecessor on coding and research evaluations, while cached-input reads cost one quarter as much. We explain when costs fall and how it differs from the more restricted Mythos.
Products & ServicesOpenAI rates Astra's cyber capabilities Critical for the first time
Critical is a capability rating, distinct from what an ordinary product permits. We examine the evaluation involving unknown vulnerabilities, its relationship to the Hugging Face incident and the reasons for restricting access.
Society & Policy