FLock adds DeepSeek V4 Flash to OpenAI-compatible API gateway

Why it matters

Jiahao Sun is extending FLock from decentralized model training into model distribution. DeepSeek can bring API volume, but FLock must show that its developer and billing tools justify displayed rates about 3.1 times DeepSeek's direct prices.

Jiahao Sun's FLock has added DeepSeek V4 Flash to the FLock API Platform, giving developers an OpenAI-compatible route to a 284-billion-parameter mixture-of-experts model with a 1-million-token context window.

FLock announced the availability on X, without specifying when the listing was added.

FLock on X

Sun, an Oxford alumnus who previously served as director of AI at Royal Bank of Canada and researched AI at Imperial College London, built FLock around a public thesis that valuable data should not have to be pooled inside a central model-training system. FLock combines federated learning, which keeps training data inside its original environment, with blockchain-based verification and incentives.

The DeepSeek listing pushes FLock further into a different part of the AI stack. FLock began as infrastructure for collaboratively training and owning models. Its API Platform, launched in September 2025, gives developers a single key for calling models that FLock trains or hosts alongside models from outside labs. DeepSeek V4 Flash belongs to the latter category.

The price of using one gateway

At publication, FLock's model catalog showed a card labeled "DeepSeek V4 Flash - Pro," with discounted pricing of $0.435 per million input tokens and $0.87 per million output tokens. Because the card combined "Flash" and "Pro" in one label, buyers still need to confirm exactly which configuration FLock serves before comparing providers.

DeepSeek publishes its own direct pricing page for V4 Flash, and FLock's displayed rates should not be treated as a like-for-like comparison until buyers confirm the served configuration and routing. The practical question for FLock is whether the gateway's workflow features can justify a paid intermediary layer for teams that could otherwise call model providers directly.

FLock is asking developers to value the layer around the model. In its API Platform Getting Started guide, FLock says users create a team to manage organizational API usage, billing and access control, and that API keys, credits, logs and members belong to that team. The same guide describes a Logs tab showing request status, latency and errors, a Usage tab tracking requests, token consumption and average cost per request, and a browser-based API Playground for testing models. FLock's API endpoint documentation identifies the endpoint as OpenAI-compatible and tells OpenAI client users to change the base URL to FLock's endpoint.

That pitch has become a distinct infrastructure market. OpenRouter markets one API across model providers. Vercel's AI Gateway gives developers unified model access and failover inside a larger developer platform. FLock has to justify its displayed rates through its developer workflow, billing system, model access and data controls.

A three-month-old model gets another distribution channel

DeepSeek released V4 Flash on April 24, 2026, more than three months before the FLock listing was observed. The official release documentation identifies V4 Flash as a 284-billion-parameter mixture-of-experts model that activates 13 billion parameters for each token. On that release page, DeepSeek described V4 Flash as "Your fast, efficient, and economical choice."

DeepSeek's model pricing page lists a 1-million-token context length for V4 Flash and V4 Pro. The April 24 release documentation separately lists a maximum output of 384,000 tokens, so the context figure describes the model's combined working window rather than the length of every response.

DeepSeek also supports thinking and non-thinking modes, JSON output and tool calls. Those features make V4 Flash relevant for coding agents that must inspect large repositories, retain long interaction histories and invoke external functions.

FLock's catalog describes the hosted model as offering "Frontier coding performance" and being built for high-volume agent workflows. The architecture and API features support long-context coding and tool use, while the speed and performance language remains FLock's product positioning. Hosting configuration, load and routing will also shape what developers experience through FLock's endpoint.

Sun expands beyond FLock's training network

FLOCK.IO LTD was incorporated on April 11, 2022. The Companies House register names Sun as an active person with significant control and records April 11, 2022, as the date he was notified. FLock's original products focused on decentralized training, validation and federated learning, allowing organizations to collaborate on models without sending raw data to a shared repository.

FLock has since built a distribution business around those models. The API Platform creates a route from model creation to paid usage. FLock also markets Moonbase-FOMO as a model-assets and distribution product, tying the API Platform to its broader distribution business. Adding models such as DeepSeek V4 Flash can attract API traffic while FLock's own training network matures.

Investors have financed that expansion. FLock announced a $6 million seed round in March 2024, co-led by Lightspeed Faction and Tagus Capital with participation from DCG, Volt Capital, OKX Ventures and Inception Capital. A $3 million strategic round, led by DCG in December 2024, brought FLock's stated total funding to $11 million. The two named rounds account for $9 million of that figure.

DeepSeek V4 Flash gives Sun a recognizable open model for selling the gateway while FLock develops its broader private-training thesis. The commercial test is whether teams find enough value in FLock's developer and billing tools to use its gateway pricing, and whether FLock makes its combined model listings clear enough for buyers to compare providers.