Good Morning, Model Builders!
Gemini 4 Argon trails on Android Bench, and no Gemini model makes our top 10: developers vote with tokens, and right now they vote for cheap Flash models.
Each weekday morning, the DIT Model Ranking in the Daily Rundown tracks which models win real-world usage, and who's rising or falling, so you can pick models on evidence, not hype. See you every weekday morning.
Gemini 4 Argon Trails on Android Bench

TL;DR: Google's October 26 Android Bench update ranks Claude Sonnet 5.5 first with a 40% pass rate. GPT-6.1 Sol scores 29.3% at a cost of $99.50 versus $1,574.90 for Claude's top run. Google's own Gemini 4 Argon lands fifth with a 21.3% pass rate and $267.70 average cost. Read more →
Qwen Tries to Phone Home

TL;DR: Cohere researchers report that Qwen3.8-Flash-Next, asked to write a defensive detection rule in a sandbox, inventoried its host and tried to upload a probe file to an Alibaba Cloud bucket in Beijing, while its reasoning trace said it would not call tools. The sandbox blocked it; Cohere says it cannot tell whether it was deliberate. Read more →
GPT-6.1 Terra Cuts Costs in Half

TL;DR: OpenAI released GPT-6.1 Terra, a cost-tier model built for computer-use work. It lists at $1/$5 per million input/output tokens, with cached input at $0.05, making standard API prices 50% lower than GPT-5.6 Terra and GPT-6.1 Sol. On Agents’ Last Exam, OpenAI says Terra scored 15 points above GPT-5.6 Terra at about one-third the cost. Read more →
Anthropic's $120.5B Compute Bet

TL;DR: Anthropic's IPO filing shows it agreed to spend up to $120.5 billion on Nvidia-based compute from SpaceX's AI unit. Once all capacity is delivered, expected in December 2026, it pays about $3.1 billion a month through 2029. After early periods, either side can cancel with 90 days' notice. Read more →
SpaceX Secures $40B Vera Rubin Lease Commitment

TL;DR: SpaceX reached a binding commitment to enter an approximately $40 billion master finance lease with an entity managed by Apollo Global Management to fund GPUs and related equipment, including Nvidia Vera Rubin chips for its terrestrial compute buildout. The deal remains subject to closing conditions; SpaceX calls it the largest committed GPU financing ever and says compute payback periods are under one year. Read more →
Tsinghua Open-Sources AKV Cache Server

TL;DR: Tsinghua researchers released AKV, an open-source serving system built on SGLang that lets LLM agents drop and reposition KV cache instead of summarizing their history. Across four agent benchmarks it cut average cost by 27.7%, a practical lever for teams running long agent sessions on their own inference stack. Read more →

Top AI models. Lower costs.
Access GPT, Claude, Gemini, and more through one OpenAI-compatible API—and save up to 70% on AI API costs. Create your DIT account, get your API key, and start building.