Good Morning!
GPT-6.1 Sol cracked the top ten yesterday, up five places to ninth, and it is the only new name in it. The model that lost the seat was MiMo v2.6 Flash, now thirteenth.
Sol's traffic is agent and code work, with nearly two thirds of its volume in agent work, and its own traffic grew almost 40% in a day. One newcomer is not the whole story, but it is the loudest sentence in it.
Model | tokens · vs. previous day (Oct 27) | ||
1 | DeepSeek V4.1 Flash · DeepSeek | 9.74T | – +16% |
2 | GLM 5.3 Flash · Z.ai | 2.43T | – +30% |
3 | GPT-6 Luna · OpenAI | 1.47T | ▲1 +33% |
4 | Hy4 preview · Tencent | 1.45T | ▼1 +7% |
5 | DeepSeek V4 Flash · DeepSeek | 1.02T | – -8% |
6 | Claude Opus 5.5 · Anthropic | 1.01T | – -8% |
7 | MiniMax M3.1 Flash · MiniMax | 0.90T | – +6% |
8 | GLM 5.3 · Z.ai | 0.64T | – -9% |
9 | GPT-6.1 Sol · OpenAI | 0.53T | ▲5 +39% |
10 | Hy3 · Tencent | 0.47T | – +0% |
Google Releases Gemini 4 Carbon

TL;DR: Google released Gemini 4 Carbon, its newest frontier model, claiming significant gains in coding, agentic tasks and cybersecurity. The company says Carbon now takes the top spot on the Artificial Analysis Intelligence Index, raising the bar for rival labs on public benchmarks. Read more →
Anthropic Details $120.5B SpaceX Compute Deal

TL;DR: Anthropic disclosed agreements worth up to $120.5 billion for Nvidia-powered compute from SpaceX's AI subsidiary through 2029. Monthly payments are expected to reach approximately $3.1 billion once all capacity is delivered, currently projected for December 2026. Either party can terminate with 90 days' notice after initial non-cancelable periods ending in early 2027. Read more →
Nvidia Opens Nemotron 3.5 Transcribe Weights

TL;DR: Nvidia released Nemotron 3.5 Transcribe, an open weight 30B mixture of experts model with about 4B active parameters. One checkpoint handles speech in 40 languages, speaker attribution, translation, contextual prompting and audio understanding, and the weights are downloadable. Read more →
Mistral Ships Responses API in Preview

TL;DR: Mistral introduced its Responses API in public preview, supporting the OpenAI Responses format for agent development. It offers structured outputs, function calling, streaming and model-dependent reasoning support. The API is stateless, requiring developers to manage conversation history themselves. Read more →
DOJ Scrutinizes Nvidia Hugging Face Deal

TL;DR: The US Justice Department is reviewing Nvidia's $12.93 billion acquisition of Hugging Face, examining whether owning the main model hub would let the chipmaker control how models reach developers and harm rivals. Read more →
SK Hynix HBM Sold Out to 2029

TL;DR: SK Hynix says its DRAM and HBM output is effectively sold out through 2029, based on existing long-term agreements and advanced customer negotiations. Signed contracts already cover more than half of its capacity for 2027 and 2028. The outlook signals sustained memory demand, though not all future production is contractually committed. Read more →
Artificial Analysis Overhauls Intelligence Index

TL;DR: Artificial Analysis revamped its Intelligence Index with new benchmarks and revised scoring, moving Kimi K3.1 Pro from fourth to ninth place. The measured gap between leading closed and open-weight models widened from three to 12 points. Version 5 now draws 60% of its scoring from private or held-out evaluations to reduce benchmark gaming. Read more →
llama.cpp Adds LiquidAI LFM3 Support

TL;DR: llama.cpp shipped version 0.8.0, adding support for LiquidAI's LFM3 family across text, vision, audio and omni models. The release also wires realtime audio into the server through new realtime and text to speech endpoints, making local speech apps easier to build. 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.