GMI Cloud offers developers a free week with three Alibaba models

Why it matters

GMI is using free access to Alibaba models to bring developers into its inference service, extending Alex Yeh's infrastructure business up to the API layer. The promotion may attract trials, but neither company has reported whether those trials become sustained usage.

Alibaba's Qwen team said in an October 9th post that a week of free access to Qwen3.8-Max, Qwen3.8-Flash and Wan3.0 was live on GMI Cloud. For GMI founder and CEO Alex Yeh, the promotion puts a new kind of product in front of developers: model access through an API, alongside the GPU infrastructure that first brought the company into the AI business.

GMI's campaign page advertises seven days of free use and describes the models as available through an OpenAI-compatible API. Its lineup presents Qwen3.8-Max for long-context work, Qwen3.8-Flash for latency-sensitive workloads and Wan3.0 for video generation. Those capability descriptions come from GMI; the promotion itself does not establish independent performance results or customer uptake.

New users must have at least $10 in their GMI account balance, according to the campaign's terms. GMI says the money is not used for the free model access and remains available for later use on its platform. That requirement adds a small step to what is otherwise a free trial. The offer page also provides a separate application path for enterprise credits. GMI says that program runs through November 5th and offers up to $10,000 per organization.

From mining sites to model APIs

Yeh's route to this offer ran through physical infrastructure. GMI started with data centers supporting Bitcoin mining, then shifted toward AI cloud infrastructure after clients and investors asked for GPU capacity. In a 2025 interview, Yeh described the next steps as moving from providing GPUs, to building out the machine-learning stack, and then to offering API access for customers who did not want to manage the hardware themselves.

The promotion follows GMI's move from GPU capacity to API access. Developers can test models without first arranging a dedicated GPU cluster, while GMI gets a chance to introduce its inference service to users who may later need more sustained capacity. This is the commercial logic of the offer, not evidence that it has generated paying customers.

GMI's 2024 Series A financing shows the capital behind the infrastructure side of Yeh's bet. The round totaled $82 million, comprising $15 million in equity and $67 million in debt, and was led by Headline Asia with participation from Banpu and Wistron, TechCrunch reported. The investors brought different assets to the deal: Banpu was to provide power, while Wistron planned to co-develop products with GMI. The financing was announced in October 2024, well before this model-access campaign.

The shift from infrastructure to API access also changes the first interaction a potential customer has with GMI. A developer can try an Alibaba model through a familiar interface before deciding whether to commit to a longer-term setup. For Yeh, who has said customers increasingly want a simple API rather than direct interaction with GPUs, that lowers the barrier to trying GMI's service. It does not establish that the trial will translate into a paid workload.

A distribution channel for both sides

The arrangement also gives Alibaba's models another route to developers. RuntimeWire previously reported that Nebius added an Alibaba Qwen model to its hosted inference menu, an example of model access spreading across infrastructure providers. In GMI's case, the offer bundles three products, including a video-generation model, into a single period of free access.

The campaign page sets out the practical terms, while Qwen's post supplies the public endorsement and points users to GMI. Alibaba's models get another distribution channel, and GMI can introduce its inference platform to developers testing them. The companies have not reported results from the promotion, so any claim about sign-ups, usage or conversion would go beyond the evidence.

Yeh built GMI by repurposing a business organized around data centers and compute. The Qwen promotion is a smaller test of whether that infrastructure can meet developers at the API layer, where trying a model takes less commitment than provisioning hardware. For GMI, a free week is an invitation to sample the service; the harder measure is whether users keep routing workloads through it after the offer ends.