Archive · 14 Aug 2026
Alibaba says new AI model Qwen3.8-Max matches Anthropic's best
Alibaba is a big technology company from China.
It made a new super-smart computer helper called Qwen3.8-Max.
The company says this helper is just as smart as the best helper made in America by a company called Anthropic.
This helper is special because it can write computer programs all by itself.
Alibaba says it worked for more than 16 days without stopping on one big project.
It even built its own tools before writing the program it was asked to make.
The helper is very big — it has 2.4 trillion parts, but only uses a small number of them at a time.
However, nobody else has tested it the same way, so we only have Alibaba's word that it is that good.
Other companies in China are also making very smart helpers.
This means China is getting really good at making computer helpers.
Alibaba unveiled Qwen3.8-Max, claiming it can compete with the best American AI systems and naming Anthropic's Claude Fable 5 specifically.
Qwen3.8-Max is a mixture-of-experts model with 2.4 trillion total parameters, about 95 billion active per inference, and a one-million-token context window.
Alibaba says the model ran autonomously for more than 16 days on a software engineering project, producing the open-sourced framework oh-my-cli.
The 16-day autonomy run is entirely vendor-reported and has not been replicated by an independent party, raising verification concerns.
Six of the top ten models on independent rankings in mid-July were Chinese, and rivals include Moonshot's Kimi K3, DeepSeek's V4 Flash, and a reported ByteDance model of up to 10 trillion parameters.
- Who
- Alibaba, the Chinese technology giant, along with other Chinese AI laboratories including Moonshot AI, DeepSeek, and ByteDance.
- What
- Unveiled Qwen3.8-Max, a 2.4-trillion-parameter mixture-of-experts model it claims matches Anthropic's Claude Fable 5 after a 16-day autonomous software engineering run.
- When
- Not explicitly stated for the Qwen3.8-Max launch; the article references mid-July rankings, the July 27 release of Kimi K3, and the July 31 release of DeepSeek V4 Flash.
- Where
- China, home of the Chinese AI laboratories behind the competing models discussed.
- Why
- To position Alibaba's model as competitive with the best American AI systems and signal growing confidence in Chinese AI.
Alibaba's capability claim
Independent verification concerns
Matching Anthropic's best
Alibaba's capability claim
Alibaba states Qwen3.8-Max can compete with the best American AI systems, naming Anthropic's Claude Fable 5, and backs the claim with a 16-day autonomous software engineering demonstration.
Independent verification concerns
The claim and the 16-day run are entirely vendor-reported, designed and run by Alibaba with no independent replication; recent cases like Meta's composite Llama 4 scores and Muse Spark 1.1's benchmark gap justify caution on self-reported numbers.
Open-source positioning vs. access model
Alibaba's capability claim
Qwen3.8-Max is positioned as an open-source platform for software development and collaborative work.
Independent verification concerns
The New Stack described the release as 'an API business model wearing an open source jacket,' citing the gap between open-weight positioning and how the model is actually accessed and monetised.
- Model
- Qwen3.8-Max
- Comparable model named
- Anthropic's Claude Fable 5
- Total parameters
- 2.4 trillion (mixture-of-experts)
- Active parameters per inference
- ~95 billion
- Context window
- 1 million tokens
- Headline demonstration
- 16+ day autonomous run producing open-sourced oh-my-cli framework from an empty repository
- Chinese models in top 10 (mid-July)
- 6 of 10, including all of the top 5
- Notable rival releases
- Kimi K3 (2.8T parameters, Jul 27), DeepSeek V4 Flash (Jul 31), ByteDance reportedly pre-training up to 10T parameters