Tencent releases Hy4 Flash with image support and a context window of up to 1 million tokens

Tencent releases Hy4 Flash with image support and a context window of up to 1 million tokens

Ondřej Barták
Ondřej Barták
Entrepreneur and Programmer
16. 11. 2026
1 minutes reading · 9 views
Tencent releases Hy4 Flash with image support and a context window of up to 1 million tokens

Tencent’s Hunyuan team has released Hy4 Flash, a lower-cost model derived from its flagship Hy4. It is designed for high-volume applications such as chat, document processing and routine code edits. The model accepts text and images and supports a context window of up to 1 million tokens.

Images, tools and code edits

According to Tencent, Flash inherits its image capabilities and long context window from Hy4. It can recognize text in images, analyze screenshots, interpret charts and parse user interfaces.

For automated computer use, the model interprets what is on the screen and selects actions in agent loops. It also supports calls to application-defined functions and external services. For programming, it targets routine edits, refactoring and lightweight code generation.

Comparisons with Hy4 and Claude Haiku 5.5

Tencent says Hy4 Flash achieved results comparable to Claude Haiku 5.5 in high-reasoning mode in blind tests where human evaluators assessed everyday tasks.

In the company’s cost-versus-performance comparison, Flash received a score of 42.1, while the flagship Hy4 scored 51.4.

Architecture with Engram memory

Hy4 Flash uses a mixture-of-experts architecture, which engages only some of its parameters in a given computation. Of its 224 billion total parameters, it activates 12 billion for each token processed.

The model includes Engram, a memory module with 64 billion parameters. Tencent says it uses table lookups to retrieve representations of recurring token patterns, reducing the computation required.

API and input reuse

Tencent lists API rates per million tokens of $0.003 for cached input, $0.070 for uncached input and $0.281 for output; the company says these are approximately one-fifteenth, one-twelfth and one-ninth of the corresponding Hy4 rates.

Discounted cached input operates separately from the Engram module inside the model. It applies to input that Tencent’s platform recognizes as eligible for reuse, such as repeated prompt prefixes, system instructions or parts of documents.

Advertisement

Content created with help from UpTier.

SEO and GEO on autopilot. UpTier’s multi-agent systems write and optimize content for search engines and AI answers.

Discover UpTier ↗

Category:AI
Did you enjoy this article?
Discover more interesting posts on our blog
Back to blog

Related posts

OpenAI introduces GPT-6.2 Astra and Sol for professional and everyday workOpenAI introduces GPT-6.2 Astra and Sol for professional and everyday work
OpenAI has introduced two models: Astra for complex professional tasks and Sol for everyday work. Both are available in ChatGPT Work, Codex and through the API.
3 min read
17. 11. 2026
Companies Linked to Chinese Spy Invested in DeepSeek and Promised Banned Nvidia ChipsCompanies Linked to Chinese Spy Invested in DeepSeek and Promised Banned Nvidia Chips
Companies linked to an alleged Chinese spy joined DeepSeek’s latest funding round. In exchange for a stake, they reportedly promised banned Nvidia Blackwell chips.
3 min read
16. 11. 2026
Long Context Is an Expensive Luxury. Nvidia Study Shows How to Make AI Agents Faster and CheaperLong Context Is an Expensive Luxury. Nvidia Study Shows How to Make AI Agents Faster and Cheaper
AI agents do not need to process their entire history at every step. An Nvidia study shows that smart context trimming can significantly speed up their work, cut costs, and preserve accuracy.
5 min read
16. 11. 2026
Přihlaste se k odběru našeho newsletteru
Zůstaňte informováni o nejnovějších příspěvcích, exkluzivních nabídkách, a aktualizacích.
CodedTrip

Operated by CodedTrip LLC, USA.

YouTube
TikTok