Closed-source Model

Open-source models generally refer to AI models whose model weights can be publicly downloaded and which allow free use and modification, as opposed to closed-source models that are only accessible via API.

GPT-4 and Claude are closed models; users can only access them via paid API calls and cannot directly obtain model parameters.

Open-source models still lag behind top closed-source models on some complex reasoning and long-context tasks.

The capabilities of open-source models are rapidly approaching those of closed-source models, and the ongoing price war is continuously compressing profit margins.

Closed-source API models can control abuse through server-side filtering, usage monitoring, and rate limiting, whereas open-source models cannot enforce such controls.

NousResearch在LLaMA、Mistral等开源基座模型之上通过微调策略打造适配Agent工作流的模型变体

Open-source models have historically lagged behind closed-source models in performance by 6 to 18 months.

Closed-source frontier models clearly outperform open-source models on the cursive handwriting trajectory recovery task, and this ranking also holds on the IAM-OnDB dataset.

For the LLM component, Reimer prefers open-weight models, with controllability as the core reason; closed frontier models have structural problems such as speed possibly declining over their lifecycle and not necessarily catering to P99 latency.

For open-weight models, researchers can inspect their training data and attention patterns, and even reproduce the generation process of specific token sequences; for closed-source models, any behavioral observation can only be based on the input-output interface, with no deep forensics possible.

A perceptible gap still exists between open-source local models and top closed-source models.

Nathan Lambert proposes that open and closed models operate on different exponential timescales: closed-source labs are highly integrated and can be first into high-value markets (such as coding), with revenue going from zero to the hundred-billion-dollar level; the open-source ecosystem develops more slowly due to poor information flow and a lack of economies of scale.

Andreessen believes the gap between open-source and closed-source models is roughly six months to a year.

On the Pareto frontier of the cost-quality chart, almost all models are closed-source, and local solutions as a whole sit inside the frontier, making them dominated choices.

Closed-source models have a floor on price cuts because they need to support commercial organizations; the cost structure of open-source models is determined by the community and hardware efficiency, so their room for price cuts is entirely different.

In industries sensitive to data sovereignty, such as finance and healthcare, closed-source solutions still command a trust premium.

Cross-cloud distribution of closed-source models is still in its early stages.

Enterprises can achieve hybrid deployment through an AI gateway, routing high-frequency simple tasks to low-cost open-source models and handing complex tasks to closed-source models.

On KEEP-Bench, the best closed-source model retained all required evidence in only 61.6% of cases.

Traffic on Vercel AI Gateway shifted from the vast majority of tokens flowing to closed-source models at the start of the year to 80% of tokens now flowing to open-source models.