Hy4

Tencent · On OpenRouter since November 4, 2026 · Pinned version id: tencent/hy4-20261104

Price

$0.6254

mixed / 1M tokens

Mixed price per 1M tokens: 3 input tokens to 1 output token.

in $0.417 · out $1.25 · Cached input $0.021

Models that think before they answer can use many more output tokens, so your real cost can be higher than the figure shown.

US dollars per million tokens, converted to your currency at the NBP rate.

Tasks, with the evidence

Each line names the measurement it rests on. A task that is not listed has no measurement or declared feature we can show.

Writing code

Measured by a third party

Code: 1,248 · Rank 41 of 122, among scored models · Design Arena, read November 21, 2026

Cost of this job: $0.25

Hard questions and reasoning

Measured by a third party

Intelligence Index: 34.6 · Rank 23 of 66, among scored models · Artificial Analysis, read November 21, 2026

Cost of this job: $0.11

Models that score at least as high for less in this scenario: 2

Websites and apps

Measured by a third party

Websites: 1,247 · Rank 44 of 127, among scored models · Design Arena, read November 21, 2026

UI components: 1,179 · Rank 66 of 118, among scored models · Design Arena, read November 21, 2026

Cost of this job: $0.13

Models that score at least as high for less in this scenario: 6

Games and 3D Top 25%

Measured by a third party

Game development: 1,268 · Rank 26 of 121, among scored models · Top 25% · Design Arena, read November 21, 2026

3D scenes: 1,237 · Rank 31 of 114, among scored models · Design Arena, read November 21, 2026

Cost of this job: $0.13

Models that score at least as high for less in this scenario: 3

Long documents

Declared, not measured

Context window: 1M tokens as declared. It says how much fits, not how well the model uses it.

Cost of this job: $0.52

Images, files, audio, video in

Declared, not measured

Accepts: images. Declared by the provider through OpenRouter; we did not test it.

Cost of this job: $0.15Text-token prices only; images, audio and video may be billed separately.

The size of each job is our assumption; the prices are real.

  • Writing code: Requests: 40. Input tokens per request: 30,000. Output tokens per request: 1,500. Share of input served from cache: 70%.
  • Hard questions and reasoning: Requests: 10. Input tokens per request: 3,000. Output tokens per request: 8,000. Share of input served from cache: 0%.
  • Websites and apps; Games and 3D: Requests: 8. Input tokens per request: 12,000. Output tokens per request: 10,000. Share of input served from cache: 30%.
  • Long documents: Requests: 3. Input tokens per request: 400,000. Output tokens per request: 4,000. Share of input served from cache: 0%.
  • Images, files, audio, video in: Requests: 15. Input tokens per request: 20,000. Output tokens per request: 3,000. Share of input served from cache: 30%.

Thinking is billed as output tokens; a model that thinks longer costs more than this estimate. Ranks are among all models we list, including older ones. The share is rounded up to whole models, so a rank slightly past the exact share still counts. Models with equal scores share the better rank.

Cheaper models with a similar benchmark score

Rule: an Intelligence Index of at least 31.6, a mixed price at least 25% lower, at least the same input types, and tool support. Lower mixed price (3 input tokens to 1 output token); check the flags and your own prompt sizes.

  1. Claude Haiku 5.5 Anthropic 68% cheaper $0.20 · Index 36.5 (this model: 34.6) Differences to check: Price rises for long prompts Compare
  2. Muse Ember Meta 68% cheaper $0.20 · Index 36.0 (this model: 34.6) Compare

Price by provider

The price used in the catalog is the vendor's own; when the vendor has no API of its own, it is the cheapest regular host.

ProviderInputOutputCached inputMixedNotes
Tencent Vendor Price used here $0.417$1.25 $0.021 $0.6254 fp4 · Numeric precision used by this host. Lower precision (fp4, int4) is cheaper and can change quality. · promo −50%
Together $0.834$2.50 $0.042 $1.25

Specifications

Context window
1M
Max output
66k
Accepts
text and images
Answers with
text
Tools
Yes
Structured output
Yes
Reasoning
Optional
Thinking effort levels: none, low, and high
Thinking is on by default
Weights on Hugging Face
Hugging Face A Hugging Face repository is linked. We did not check the licence; read it before you build on it.
Knowledge cutoff
No data
On OpenRouter since
November 4, 2026
Price check
one source

As listed by OpenRouter. "Not listed" means OpenRouter does not list it for this model.

Benchmark scores

Scores come from third parties and are relayed by OpenRouter. We do not run the tests. Higher is better.

SourceScoreRank among scored
Intelligence Index
Artificial Analysis, read November 21, 2026
34.6Rank 23 of 66, among scored models
Websites
Design Arena, read November 21, 2026
1,247Rank 44 of 127, among scored models
UI components
Design Arena, read November 21, 2026
1,179Rank 66 of 118, among scored models
Game development
Design Arena, read November 21, 2026
1,268Rank 26 of 121, among scored models Top 25%
3D scenes
Design Arena, read November 21, 2026
1,237Rank 31 of 114, among scored models
Code
Design Arena, read November 21, 2026
1,248Rank 41 of 122, among scored models
agents/pdf
Design Arena, read November 21, 2026
1,219Rating 1,219; rank 15 at the source. Too few scored models for a tier.

Intelligence, Coding and Agentic indices: Artificial Analysis (artificialanalysis.ai). Design categories: Design Arena (designarena.ai). Both are relayed by the OpenRouter public model list and were read on November 21, 2026. We compute ranks, tiers and costs from these numbers; the sources have not reviewed them. Scores belong to the model, not to the route it is served on.

Artificial Analysis · Design Arena · Relayed by OpenRouter. opens in a new tab

Sources and dates

Prices: OpenRouter routes and a second price table, checked November 21, 2026. Specifications: OpenRouter, November 21, 2026. Exchange rate: NBP, November 20, 2026. Benchmarks: Artificial Analysis and Design Arena via OpenRouter.