Alibaba previews Qwen3.8 with 2.4 trillion parameters, claims second place without benchmark data
Alibaba unveiled Qwen3.8 at the World Artificial Intelligence Conference in Shanghai, claiming the 2.4 trillion parameter model ranks second only to Anthropic's Fable 5. The company provided no benchmark scores, model card, or independent verification to support the claim.
Alibaba previews Qwen3.8 with 2.4 trillion parameters, claims second place without benchmark data
Alibaba unveiled Qwen3.8 at the World Artificial Intelligence Conference in Shanghai, claiming the 2.4 trillion parameter model ranks second only to Anthropic's Fable 5. The company provided no benchmark scores, model card, or independent verification to support the claim.
Qwen3.8 is Alibaba's first multimodal model above one trillion parameters, handling images, video, documents, and text. According to Alibaba, the model should outperform its previous flagship Qwen3.7-Max, particularly at coding and office tasks including full-stack development and data analysis. Lead developer Shuai Bai said the model's understanding matches or beats leading proprietary systems, though no evidence was provided.
The model is available now through Alibaba's Token Plan subscription and Qoder developer tools at 10% of standard pricing during the trial period. Alibaba promises to release full open weights "soon" but has not specified a date or license terms.
Missing verification
Alibaba shipped Qwen3.8 without benchmark scores, a model card, or any independent testing. This marks a departure from the company's previous release: Qwen3.7-Max launched in May with published results across standard benchmarks.
No public leaderboard has scored Qwen3.8. On LMArena, where Fable 5 ranks first, the older Qwen3.7-Max sits well below the top positions. Without released weights, external researchers cannot verify Alibaba's claims.
Chinese model competition intensifies
The announcement came three days after Moonshot released Kimi K3, a 2.8 trillion parameter open model that topped a major coding leaderboard. Multiple Chinese labs are now shipping frontier-scale models: Moonshot has Kimi K3, Zhipu has GLM-5.2, and most are releasing weights publicly.
Both Alibaba and Moonshot are deploying models in paid products before releasing weights, converting launch momentum into customers before open access becomes available. This represents a shift for Alibaba, which has historically kept its largest models closed.
Alibaba shares rose as much as 5.4% on Monday following the announcement.
What this means
Alibaba's announcement signals escalating competition among Chinese AI labs, with multiple companies now claiming frontier-scale capabilities and promising open weights. The lack of benchmarks or verification data represents either rushed competitive pressure or a calculated marketing move. If Alibaba releases weights matching its claims, it would push the open-weight frontier significantly forward. Until then, the "second only to Fable 5" claim remains unverified marketing. The pattern of paid deployment before weight release suggests Chinese labs are developing sustainable business models around open-weight releases rather than purely research-driven distribution.
Related Articles
MiniMax releases M3.1-Flash: 428B-parameter multimodal MoE with 1M-token context and weights on Hugging Face
MiniMax has published MiniMax-M3.1-Flash on Hugging Face: a native multimodal mixture-of-experts model with about 428B total parameters, 23B activated per token, and a context window of up to 1M tokens. It builds on MiniMax-M3 with sparse attention, quantization-aware training and DSpark speculative decoding.
Xiaomi open-sources Xiaomi-Lens-0, a 4B multimodal model that outputs pixel-level mask tokens mid-answer
Xiaomi's SeerRay-Lab has open-sourced Xiaomi-Lens-0, a 4B multimodal model built on Qwen3.5-4B. The model emits mask tokens for pixel-level image regions in the middle of an answer. The lab also released a benchmark called LensBench.
OrcaRouter releases OrcaCyber Nano 1.0, a 27B local security model with 262K context and GGUF builds
OrcaRouter has published OrcaCyber Nano 1.0, a 27B vision-language reasoning model for defensive security, as GGUF files for local inference with llama.cpp. It supports a 262,144-token context window and text, image and video input. Benchmark scores are self-reported, and the harness details are not yet published.
Tencent releases WeVisOne-VL-4B, a Qwen3-VL-based model for object detection, segmentation and alpha matting
Tencent's WeChat Vision team released WeVisOne-VL-4B, a vision-language model built on Qwen3-VL that handles object-level and pixel-level perception. It adds a 9.6M-parameter pixel decoder to produce segmentation masks and full-resolution alpha mattes.
Comments
Loading...