Good Morning!
OpenRouter's top-ten text models moved 29.7 trillion tokens yesterday, 6.0% below the same weekday a week earlier. Tencent's Hy4 jumped 78.6%, while DeepSeek V4.1 Flash held first at 10.9 trillion tokens despite a 17.5% pullback.
Model | tokens | WoW (11/3) | |
1 | DeepSeek V4.1 Flash · DeepSeek | 10.92T | – -17% |
2 | Hy4 · Tencent | 2.83T | ▲1 +79% |
3 | GLM 5.3 Flash · Z.ai | 2.10T | ▼1 -29% |
4 | Claude Opus 5.5 · Anthropic | 1.11T | ▲1 +1% |
5 | GPT-6 Luna · OpenAI | 1.07T | ▼1 -5% |
6 | DeepSeek V4 Flash · DeepSeek | 0.78T | – -13% |
7 | GPT-6.1 Sol · OpenAI | 0.67T | ▲3 +32% |
8 | MiMo-V2.6-Flash · Xiaomi | 0.66T | ▲10 +85% |
9 | GLM 5.3 · Z.ai | 0.60T | ▼1 +14% |
10 | Nemotron 3 Ultra · NVIDIA | 0.58T | ▲1 +16% |
Gemini 4 Carbon Opens Paid API

TL;DR: Google opened Gemini 4 Carbon to paying developers. The Gemini API charges $2 per million input tokens and $10 per million output tokens through Nov. 30, then $4 and $20. Google also previewed /swarm in Antigravity, which lets one prompt coordinate multiple agents. Read more →
MiniMax Opens H3-Image

TL;DR: MiniMax released H3-Image weights on Hugging Face. The model generates and edits images at about 16 megapixels in six sampling steps and accepts up to 16 reference images. Its community license excludes the US, UK, EU and South Korea. Read more →
ERNIE 5.5 Flash Preview Goes Live

TL;DR: Baidu’s ERNIE 5.5 Flash Preview is now available through Qianfan with a 1 million token context window. OpenRouter lists it at $0.28 per million input tokens, $1.13 per million output tokens and $0.028 per million cached input tokens. Read more →
LG Opens Free K-EXAONE API

TL;DR: LG AI Research opened the K-EXAONE 2.5 API for free noncommercial use. Developers can get a key through the EXAONE API Platform. In LG’s own tests, the model scored 78.0 on BrowseComp and 59.2 on SWE-bench Pro. Read more →
Nota Runs Kimi on 8 B200s

TL;DR: Nota removed 37.5% of Kimi K3.1 Pro’s routed experts and ran the smaller model on eight Nvidia B200 GPUs with a 1 million token context window. Nota reports a GPQA Diamond score of 93.43, compared with 94.20 in Moonshot’s original model card. Read more →
Runpod Batch Up to 30% Off

TL;DR: Runpod made Batch Inference generally available on Oct. 28. It runs jobs that can wait on separate workers behind the same Serverless endpoint. Discounts reach 30% versus on-demand rates, depending on the GPU. H100 and B200 batches get 20% off. Read more →
Zyphra Opens TTS Weights

TL;DR: Zyphra released ZONOS3-Mini under Apache 2.0. The text-to-speech model has 100 million active parameters out of 800 million total. It can clone voices from a sample and generate speech in real time in the browser via WebGPU. 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.