Qwen 4
PulseAugur coverage of Qwen 4 — every cluster mentioning Qwen 4 across labs, papers, and developer communities, ranked by signal.
- 2026-10-06 product_launch A partner of Alibaba indicated that Qwen 4 is planned for release by the end of October. source
- 2026-10-02 research_milestone Alibaba's Qwen team is training the next iteration of their large language model, Qwen 4. source
- 2026-08-23 research_milestone A leaked model, potentially Qwen 4 and codenamed "korrine", is being tested and is expected to match the performance of Fable 5 and Mythos 5. source
4 day(s) with sentiment data
-
Alibaba's Qwen models to reach 10T parameters, eyes AI-driven development
Alibaba Group's Qwen models are set to scale significantly, with plans for Qwen4, Qwen4.5, and Qwen5 to reach parameter counts between 5T and 10T, building on the 2.4T parameters of Qwen3.8. The company is also focusing…
-
Alibaba partner suggests Qwen 4 release by end of October
A partner of Alibaba has indicated that Qwen 4 is anticipated for release by the end of October. This timeline appears to be sooner than expected, given the previous release cadence between Qwen 3.6 and 3.8. While speci…
-
Alibaba's Qwen 4 Large Language Model in Training, Architecture Available
Alibaba's Qwen team is currently training Qwen 4, the next iteration of their large language model. While the full model is still under development, the architecture it's based on, Flash-Next, is already available and c…
-
Cursor AI assistant users report new "condor" model, possibly Qwen 4
Users of the Cursor AI coding assistant have observed a new model appearing under the codename "condor." When queried, this model identified itself as Qwen, leading to speculation that Cursor may be testing Qwen 4. This…
-
Qwen 4 and GLM 6 show strong performance in Agent Arena benchmark
The Agent Arena benchmark has shown promising preliminary results for the Qwen 4 and General Language Model (GLM) 6 models. These open-weight models are demonstrating performance that may rival established models like M…
-
llama.cpp adds TENSOR_READ_LAZY for efficient model loading
A pull request to the llama.cpp project introduces a new feature called TENSOR_READ_LAZY, developed by ngxson. This enhancement aims to improve the efficiency of loading large language models by allowing engrams for mod…
-
Alibaba's Qwen team announces new architecture support with TokenSpeed
Alibaba's Qwen team announced support for their new architecture, including GDN + QSA, N-gram embedding, and FP8 precision. This support was provided by lightseekorg, which offered day-0 integration for TokenSpeed. The …
-
Leaked Qwen 4 Model "korrine" Targets Fable 5, Mythos 5 Performance
A leaked model, reportedly named "korrine" and tested on Code Arena, is believed to be Qwen 4. This model is anticipated to achieve performance levels comparable to Fable 5 and Mythos 5.