Updates on Chinese AI models and their practical impact on API integration, model selection, and cost.
MarkTechPost reported on October 26, 2026 that the vLLM Semantic Router team released d3-ultra, a 27B multimodal decision model under Apache-2.0. It does not generate text: a single read of the input returns a probability for each candidate answer to every question, covering routing, classification, bounding boxes and video checks.
According to a MarkTechPost report dated October 26, 2026, Baidu released Unlimited OCR 2, a 3.34B-parameter model whose weights are published under the MIT license. It uses SR-SWA to compress attention onto a fixed reference set, supporting parsing of hundreds of pages in a single request, and reports 77.1 on MarathonDocBench.
According to a Newzhiyuan report, the National University of Singapore's xML Lab proposes Jeva, which uses Qwen3.5-4B as its backbone plus a decision block that directly outputs a choice, null, or score; the paper reports zero order-flip rate and 69.88% accuracy with 2,048 candidates, and the weights have been released on Hugging Face.
According to a MarkTechPost report dated October 25, 2026, BOSS Zhipin's Nanbeige LLM Lab released the 33B dense model Nanbeige4.3-33B on Hugging Face under the Apache 2.0 license, along with a Base version for custom post-training. It supports a 524,288-token context and a self-reported table leads on 17 of 20 benchmarks, but some competitor figures conflict with those rivals' official model cards.
According to a Xinzhiyuan report on October 24, 2026, Kimi K3.1 Pro scored 55 on the Artificial Analysis Intelligence Index, ranking fourth globally and 11 points above its predecessor; Terminal-Bench 4.0 rose from 13% to 65.7%, and a Highspeed version with 3x faster output was also released.
According to an October 23, 2026 report by MarkTechPost, Kyutai released CADRE, a causal diffusion representation autoencoder; the paper claims its world model runs at 20 fps on a single H100 and responds within 185 to 216 milliseconds. The code is MIT, but the Hugging Face weights are CC BY-NC 4.0, excluding commercial use, and the source also lists shortcomings such as robotic manipulation action following.
According to a Marktechpost report on October 23, 2026, Moonshot AI released Kimi K3.1 Pro, the first model in the K3.1 series, using a 2.8T total-parameter, 104B activated-parameter MoE structure, supporting a 1M-token context, and now live on the Kimi API, with API pricing lower than K3 and open weights expected in November 2026. Its self-reported benchmarks are near the frontier on coding, but clearly behind on scientific terminal tasks.
According to an October 23, 2026 report by The Decoder, Moonshot AI released Kimi K3.1 Pro, which scored 55 on the Artificial Analysis Intelligence Index, 11 points higher than K3, ranking fourth. The report also says the standard version's per-token price was cut by about 33%, that a Highspeed variant exists, and that open weights are planned for release by November 4.
According to a Leiphone report on October 23, 2026, an official Kimi teaser contained an easter egg for "Kimi K3.1"; developers also spotted a kimi-k3-1 identifier in Moonshot's API configuration and a K3.1 option added to Kimi Code's model list; an SGLang merge request was said to add a Medium thinking tier, expanding thinking strength from three tiers to four. AI blogger Chris claimed about a week of testing and a cost around 70% of K3, and the report also said its 3D cost is about 1/14 of GPT-6 Astra. All of these are rumors and single-party testing.
MarkTechPost reported on October 22, 2026 that Gradium open-sourced Phonon, an on-device text-to-speech family. The English Phonon 31m has 31M parameters, 64.5 MB q8 weights, and an Apache 2.0 license. Gradium's internal tests put its WER at 1.70% on a difficult sentence set and time to first audio at 6.9 ms on a MacBook Air M5 CPU.
According to an Allen AI blog post published on October 22, 2026, OlmoEarth v1.3 is a satellite embedding model that reads annual time series from Sentinel-1, Sentinel-2 and Landsat, producing a 128-dimensional vector for every 10-meter pixel, with the first 64 dimensions usable on their own. The team claims its 128-dimensional embeddings scored highest on 20 of 27 evaluations, and is releasing global precomputed embeddings, weights, pretraining data and code under the OlmoEarth Artifact License.
According to a Leiphone report dated October 22, 2026, Volcano Engine has started the first batch of enterprise trials for the Seedance 2.5 4K model, tied to projects and directed to 20 studios and directors worldwide. The report says the version emphasizes character performance, reference generation and editing consistency, and one trial studio said it can first generate a draft at 480P and then regenerate at 4K after approval.