Moonshot AI and Alibaba release 2.8T and 2.4T parameter models, claim performance near GPT-5.6 and Claude Fable 5
Within days, Moonshot AI and Alibaba unveiled what they claim are frontier-class models. Moonshot's Kimi K3, at 2.8 trillion parameters, and Alibaba's Qwen3.8, at 2.4 trillion parameters, will both be released as open-weight models with full weights available for download.
Moonshot AI and Alibaba Release Massive Open-Weight Models
Moonshot AI and Alibaba have announced two large-scale AI models they claim can compete with OpenAI's GPT-5.6 Sol and Anthropic's Claude Fable 5, with both companies committing to open-weight releases.
Kimi K3: 2.8 Trillion Parameters
Beijing-based Moonshot AI released Kimi K3 on July 19, 2026, describing it as the world's largest open-source AI system at 2.8 trillion parameters. According to Moonshot's internal testing, K3 ranks above nearly every US model except GPT-5.6 Sol and Claude Fable 5, though it outperformed both on certain unspecified benchmarks. Full model weights will be released July 27, 2026.
Qwen3.8: 2.4 Trillion Parameters
Alibaba followed with a preview of Qwen3.8 over the weekend, claiming it is "one of the most powerful models available today" and "second only to Fable 5." The 2.4 trillion parameter model is described as "continuously evolving" and will be released as open-weight "soon," with no specific date provided.
Pricing Not Yet Disclosed
Neither company has announced pricing for API access to these models. Context window sizes, benchmark scores, and other technical specifications remain undisclosed pending full release.
Open-Weight vs. Proprietary
Both Chinese companies are emphasizing their commitment to making model weights publicly available for download and modification, contrasting with OpenAI and Anthropic's proprietary approach. Neither US company discloses parameter counts for their frontier models.
The releases follow DeepSeek's low-cost model announcement last year, which similarly claimed performance competitive with leading US systems.
US Export Controls Context
The announcements come as Washington restricts China's access to advanced chips through export controls. The US government recently forced Anthropic to pull its most capable system from certain markets over concerns about foreign competitors.
What This Means
These releases represent the most significant challenge to US AI labs' frontier position since DeepSeek. The 2.8T and 2.4T parameter counts substantially exceed most publicly known model sizes, though parameter count alone doesn't determine capability. Independent benchmarking will be critical once full weights are released July 27th and beyond. The open-weight strategy directly challenges the proprietary approach of OpenAI and Anthropic, potentially accelerating global AI development outside US control. If the performance claims hold up under independent testing, these models could shift the competitive landscape significantly, particularly for developers seeking alternatives to expensive API-based services.
Related Articles
NVIDIA releases Nemotron 3.5 Super VL NVFP4: 120B-parameter, 12B-active vision-language model with 1M context
NVIDIA has published Nemotron 3.5 Super VL in NVFP4 quantization on Hugging Face: a 120B-parameter, 12B-active hybrid Mamba-2/Transformer MoE with a 1M-token context window and text, image and video input. NVIDIA reports the quantized checkpoint stays close to BF16 on most benchmarks. It fits on one DGX Spark, one H200, two H100s or one GB200.
Tencent releases Hy-Memory-30B-A3B, a 3B-active model that writes agent memory for Mem0 and MemOS
Tencent Hunyuan released Hy-Memory-30B-A3B, a 30B-parameter mixture-of-experts model with about 3B active parameters, built to run the write path in external memory frameworks. Tencent reports a 55.49 cross-harness average, ahead of Claude Opus 5 (54.62) and DeepSeek V4.1 Flash (53.64) on its own fixed 20% benchmark subsamples.
DeepSeek V4.1 Flash 1017 launches with 552B CED backbone, 1M context, $0.15/$0.60 per 1M tokens
DeepSeek has released V4.1 Flash 1017, an October 17, 2026 snapshot built on its new Causal Encoder-Decoder architecture. It uses a 552B-parameter sparse MoE backbone with a 1M-token context window, native image understanding, and pricing of $0.15 input / $0.60 output per 1M tokens on OpenRouter.
Elorian launches Glimpse, a visual reasoning model priced at $0.60/$3 per 1M tokens
Elorian, the Palo Alto startup that raised a $55M seed at a $300M valuation, has released Glimpse as a research preview with API access at $0.60 per 1M input tokens and $3 per 1M output tokens. The model can crop, pull video frames, and run code mid-reasoning. Benchmark claims come from Elorian's own evaluations.
Comments
Loading...