Latest updates
What's new in LLMrank - the models we measure, how honest our reports are, and how we help you follow your visibility over time.
-
Google DeepMind released Gemini 4 Carbon, which currently ranks first on the Artificial Analysis Intelligence Index and leads several cybersecurity and coding benchmarks. Artificial Analysis launched Intelligence Index v5, utilizing private test sets to mitigate data contamination, with Claude Opus 5.5 previously leading the index. Elorian introduced the DecoyCount benchmark to evaluate visual reasoning and object counting in models like Glimpse.
Sources: news@daily.therundown.ai, news@alphasignal.ai, lspace@mail.beehiiv.com
-
Perplexity, Elastic, and Turbopuffer reported state-of-the-art retrieval accuracy using NVIDIA Nemotron 3.5 Embed, Rerank, and Parse models. New Generative Engine Optimization (GEO) strategies emphasize original research, consistent descriptors, and strategic backlinks to increase brand visibility in AI responses.
Sources: huggingface.co, dan@tldrnewsletter.com, contact@womenintechseo.com
-
Gemini 4 Argon has established a new performance lead in agentic workflows, debuting at the top of the APEX-Agents leaderboard for professional multi-step tasks. This shift toward agentic evaluation is mirrored by Anthropic's decision to disconnect its internal evaluation environments from the internet to prevent AI agents from exhibiting unintended behaviors during testing. New benchmarking trends are emerging around structured output and low-cost model efficiency. Jev has emerged as the industry benchmark for a new category of 'decision models' designed for typed outputs like probabilities and scores. Meanwhile, comparative performance tests between OpenAI's GPT-6 Luna and Anthropic's Claude Haiku 5.5 have revealed significant discrepancies in code execution accuracy, highlighting reliability gaps in the latest low-cost model tier.
Sources: theneuron@newsletter.theneurondaily.com, swyx+ainews@substack.com, theverge.com, stayingahead@mail.beehiiv.com
-
The AI search landscape is shifting toward a cycle of closed content as newsrooms and corporate sectors increasingly block AI bots to protect proprietary data. This fragmentation is forcing a pivot in AEO strategies, with new analysis revealing that Claude's citation patterns vary significantly by query type, necessitating tailored content strategies to maintain visibility. Simultaneously, the emergence of GPT-6 is introducing intelligent UI elements to ChatGPT, further evolving the interface through which users interact with generative search. Operational focus has moved toward the technical barriers of AI visibility, specifically the negative impact of gated content on LLM crawlers and the critical role of localized E-E-A-T signals. In the service sector, agencies are now formalizing the commercialization of AI search, developing specific packaging and pricing models for AI search services. Practical application is also maturing, as evidenced by the implementation of multi-step AI SEO workflows to drive organic revenue.
Sources: dan@tldrnewsletter.com, newsletters@searchengineland.com, hello@8020ai.co, contact@womenintechseo.com
-
Meta released the Muse Wonder model family, with the Full version ranking second in intelligence behind Claude Opus 5.5 and outperforming Claude Fable 5.1 and GPT-6 Astra in independent rankings for coding and legal tasks. Artificial Analysis released the AA-3D benchmark, identifying GPT-6 Sol as the most cost-effective top-performing model.
Sources: newsletter@genai.works, the-superpower@mail.beehiiv.com, huggingface.co, news@daily.therundown.ai
-
Elorian and Scale AI released the Humanity’s Sixth Sense benchmark to measure intuitive visual reasoning, where GPT 6 Astra and other frontier models lag behind human performance. Meta's Muse Wonder model demonstrated high performance on the MemoryBench benchmark, showing results comparable to GPT-6 Astra and Fable 5.1. Mistral Large 4 preview performance was evaluated using Artificial Analysis benchmarks.
Sources: importai@substack.com, techcrunch.com, simonw@substack.com
-
GPT-6.1 Sol demonstrated competitive performance against Claude Opus 5.5 and Astra in comparative analysis while operating at a significantly lower cost. The EU AI Office's enforcement of the AI Act is being evaluated alongside the UK AI Security Institute, emphasizing the role of third-party evaluations for model safety.
Sources: artificialintelligenceact@substack.com, swyx+ainews@substack.com
-
GPT-6.1 Sol demonstrated competitive performance against Claude Opus 5.5 and Astra in comparative analysis while operating at a significantly lower cost. The EU AI Office's enforcement of the AI Act is being evaluated alongside the UK AI Security Institute, emphasizing the role of third-party evaluations for model safety.
Sources: artificialintelligenceact@substack.com, swyx+ainews@substack.com
-
Claude Sonnet 5.5 released with performance approaching Claude Opus 5.5 in coding and image understanding. Benchmarks include SWE-bench Pro, Terminal-Bench, and web page polishing tests. GPT-6 Sol and Claude Opus 5.5 were compared via Tetris build task workload costs.
Sources: theoutput@datacamp.com, bensbites.com, stayingahead@mail.beehiiv.com, news@daily.therundown.ai
-
Pro measures 26 models, including earlier generations
Pro analyses cover both the newest models and a set of earlier ones, spanning several years of model generations. Because older and newer models learned from the web at different times, comparing them shows whether your brand's standing has been improving or standing still.
-
Refreshed model set: newest versions across every provider
Every analysis now runs against the current generation of each major AI model. When a provider ships a successor, we move to it, so what you see reflects the models people are actually using this month rather than the ones that were current when you signed up.
-
New frontier models on Pro: Claude Fable 5 and Claude Sonnet 5
Pro analyses now include the newest frontier AI models, so your visibility is measured against what people actually ask today.
-
Honest recognition, not just a name-drop
Reports now show whether an AI model genuinely recognises your brand, not simply whether it repeats the name. You see the difference clearly.
-
Two clear numbers: AI Visibility and AI Readiness
Your score is now split in two. AI Visibility tells you whether AI knows and mentions you. AI Readiness tells you whether your site is technically prepared for AI to find and quote you.
-
Follow your visibility over time (Pro)
Pro domains are re-measured over time, so you can track how AI's view of you changes and see whether your work is moving the needle.
-
Clearer, more useful reports
We keep sharpening how results are presented, so the report is easier to read and points more directly at what to improve next.
See how AI describes your company today.
Check your visibility