Tencent launches Hy3-preview, a non‑reasoning LLM priced at $0.20 per 1K tokens

Tencent’s latest move is Hy3-preview, a non‑reasoning large language model that entered the market on July 21 via AtlasCloud. The model supports a 262K token context window and is priced at $0.20 for input and $0.80 for output per 1,000 tokens, positioning it as one of the most affordable options in the recent wave of releases. For developers who are still learning the basics of artificial intelligence, Hy3-preview provides a straightforward interface that emphasizes speed over deep reasoning.

The timing aligns with a broader trend: according to the AI Release Tracker, the monthly cadence of major AI releases has roughly quadrupled since 2023, and today’s influx includes models from MoonshotAI, Meta, and OpenAI. This surge is tracked by aireleasetracker.com, which shows how quickly new capabilities reach the community. Hy3-preview’s launch adds another data point to that timeline, illustrating how providers compete on price and context size.

Unlike reasoning models such as OpenAI’s GPT‑Red—highlighted in a recent ibtimes.sg report for its automated vulnerability testing—Hy3-preview is explicitly non‑reasoning, meaning it does not generate step‑by‑step thought chains. This design choice makes it ideal for tasks where raw generation speed matters more than explainability, such as drafting code snippets or generating narrative text. A quick artificial intelligence review of the model shows it excels at high‑throughput scenarios but lacks the safety scaffolding built into models like GPT‑Red.

The pricing strategy reflects a growing segment of the market that favors per‑token economics. Recent listings on pricepertoken.com show competitors like LFM2.5‑8B‑A1B at $0.03 in / $0.12 out and KAT Coder Air v2.5 at $0.15 in / $0.60 out, underscoring a price war that is reshaping how developers budget for inference. Hy3-preview sits in the middle, offering a balance of cost and context that may appeal to teams building large‑scale pipelines.

For practitioners, the decision to adopt a non‑reasoning model hinges on the specific workload. In contexts like artificial intelligence for a simple game, the model’s speed can outweigh the need for complex reasoning, delivering immediate results without the latency of deeper inference. However, safety‑critical applications will likely still rely on models that incorporate automated red‑teaming, as described in the ibtimes.sg coverage of GPT‑Red.

The market reaction suggests Hy3-preview could influence future releases, prompting other labs to consider leaner, cheaper alternatives. Will this shift the balance toward cheaper, faster models, or will the demand for safety and reasoning capabilities keep premium pricing viable? The answer will become clearer as developers test Hy3-preview in production environments.

FAQ
- What is Hy3-preview?
Hy3-preview is Tencent’s newly released, non‑reasoning large language model that offers a 262K token context window at $0.20 in / $0.80 out per 1K tokens via AtlasCloud.
- How does its pricing compare to other recent models?
Hy3-preview sits between low‑cost options like LFM2.5‑8B‑A1B ($0.03/$0.12) and higher‑priced reasoning models such as GPT‑5.6 Luna ($1.00/$6.00), providing a mid‑range choice for many use cases.
- What are the main limitations of Hy3-preview?
Because it is non‑reasoning, it lacks step‑by‑step reasoning and the built‑in safety features that models like GPT‑Red provide, making it less suitable for high‑risk applications.
- When can developers start using Hy3-preview?
The model is already available as a preview on AtlasCloud, allowing early access for developers who want to experiment with its speed and cost characteristics.