Tencent Hunyuan Hy4 Flash: Free Access, $0.070/$0.281 API Pricing & DeepSeek Comparison

Announced on November 16, 2026, Tencent Hunyuan Hy4 Flash (hy4-flash) is a 224B-total / 12B-active MoE model featuring 64B Engram conditional memory, 1M context (960K input / 64K output), multimodal image-audio-video input, and >150 tok/s decoding. Developers can access Hy4 Flash free through November 30, 2026 via WorkBuddy and OpenCode Zen (hy4-flash-free), or via Tencent Cloud TokenHub at 50% off the standard ¥0.50 ($0.070) input / ¥2.00 ($0.281) output / ¥0.02 ($0.003) cached input per 1M tokens—roughly 2.1x cheaper than DeepSeek V4.1 Flash off-peak direct API pricing.

How to use Hy4 Flash for free (Nov 16–30, 2026 promos)

1 · WorkBuddy Free QuotaFree daily Hy4 Flash quota inside Tencent’s WorkBuddy client from Nov 16 through Nov 30, 2026 (23:59 BJT).
2 · OpenCode Zen EndpointSelect model ID hy4-flash-free on OpenCode Zen for two weeks free from Nov 16 (per-IP limits apply).
3 · TokenHub 50% Launch OffHalf-price metered API billing on Tencent Cloud TokenHub until Nov 30, 2026 (12:00 BJT).

Unlike promotional waitlist drops, all three launch channels opened immediately on November 16, 2026. If you are prototyping high-volume data analysis or multimodal extraction pipelines, test on OpenCode Zen (hy4-flash-free) for zero-cost CLI runs before wiring production keys into Tencent Cloud TokenHub. Note that free community endpoints enforce per-IP throttling and should not carry regulated customer PII. Tencent Cloud Hunyuan overview · OpenCode Zen models.

Hy4 Flash vs DeepSeek V4.1 Flash: official pricing & specs

The table below compares Tencent Hunyuan Hy4 Flash against DeepSeek-V4.1-Flash-1017 across token rates, context window, multimodality, licensing, and measured throughput.

DimensionTencent Hunyuan Hy4 FlashDeepSeek V4.1 Flash (1017)
Standard Input Price¥0.50 ($0.070) / 1M tokens
(50% off on TokenHub until Nov 30, 12:00 BJT)
$0.150 / 1M off-peak direct API
($0.300 / 1M standard daytime)
Cached Input Price¥0.02 ($0.003) / 1M tokens$0.003 – $0.006 / 1M tokens
Output Price¥2.00 ($0.281) / 1M tokens
(50% off on TokenHub until Nov 30, 12:00 BJT)
$0.600 / 1M off-peak direct API
($1.200 / 1M standard daytime)
Architecture & Active Params224B total / 12B active MoE (64B Engram conditional memory + LatentMoE + 4-bit + DFly)Sparse MoE Flash architecture optimized for fast reasoning & code
Context & Token Limits1M context (960K max input / 64K max output on Tencent Cloud)1M context window
Multimodal SupportImage, audio, and video input → text outputText + standalone /v1/tools/ocr & /v1/tools/tts endpoints
Decoding Speed>150 tok/s (157 tok/s reported in third-party analytics runs)High-speed Flash tier
Weights & Self-HostingHosted-only at Nov 16 launch (open-weight release planned)MIT open-weight (downloadable & self-hostable)

DeepSeek V4.1 Flash pricing and tool endpoint details: DeepSeek OCR, TTS & Search API Pricing (2026) · DeepSeek platform overview.

Where Hy4 Flash wins—and where benchmarks trail

1. Ultra-low-cost multimodal analytics: Early independent evaluation on PlotlyDataAnalyticsBench recorded Hy4 Flash scoring 39/43 for $0.13 total run cost at 157 tok/s, compared against DeepSeek scoring 36/43 for $0.46. Combining 960K input capacity with $0.070/1M uncached input makes long-document synthesis, chart data parsing, and audio/video-to-text summarization significantly cheaper on hosted APIs.

2. Caveat on terminal & multi-file coding agents: Hy4 Flash does not sweep every agentic coding suite. Reported scores on ProgramBench (16.0) and Terminal-Bench 4.0 (19.2) lag dedicated coding models. Teams running autonomous multi-file refactors or shell-heavy CI repair should verify pass rates on a bounded repo task before migrating off DeepSeek V4.1 Flash or frontier coding tiers.

3. Hosted availability vs MIT open weights: DeepSeek V4.1 Flash already ships MIT weights for private VPC or on-premise inference. Hy4 Flash launched as a hosted API and client-integrated service first; regulated workloads requiring strict data residency outside Tencent Cloud must wait for Tencent’s planned open-weight release.

Decision checklist: Hy4 Flash vs DeepSeek V4.1 Flash

Pick Hy4 Flash when: your workload summarizes long multimodal inputs (images, audio, or video alongside 100K–960K text prompts), runs high-volume data analytics, or benefits from free November 2026 prototyping on WorkBuddy / OpenCode Zen before locking into $0.070/$0.281 API rates.

Pick DeepSeek V4.1 Flash when: you need downloadable MIT weights for self-hosting today, depend on DeepSeek’s standalone /v1/tools/ocr ($0.225/1K pages) or /v1/tools/search ($0.75/1K calls) endpoints, or run heavy terminal-based software engineering loops where ProgramBench and Terminal-Bench fidelity matter most.

Compare more options in our Best AI Coding Assistants 2026 guide or Gemini 4 Hydrogen vs Claude Sonnet 5.5.

Frequently asked questions

How can developers use Tencent Hunyuan Hy4 Flash for free right now?

Starting November 16, 2026, WorkBuddy offers a free daily Hy4 Flash quota through November 30, 2026 (23:59 BJT), and OpenCode Zen exposes hy4-flash-free free for two weeks subject to per-IP rate limits.

What is Hunyuan Hy4 Flash standard API pricing in USD and RMB?

Tencent Cloud’s standard schedule is ¥0.50 ($0.070) per 1M input tokens, ¥2.00 ($0.281) per 1M output tokens, and ¥0.02 ($0.003) per 1M cached input tokens, with a 50% launch discount on TokenHub until November 30, 2026 (12:00 BJT).

Is Hy4 Flash cheaper than DeepSeek V4.1 Flash?

On hosted token rates, yes: $0.070/$0.281 per 1M tokens is roughly 2.1x cheaper than DeepSeek V4.1 Flash off-peak direct API pricing ($0.150/$0.600) and ~4.3x cheaper than standard daytime pricing ($0.300/$1.200). DeepSeek retains an edge on MIT self-hosting and terminal/software-engineering benchmarks.

Can I download or self-host Hunyuan Hy4 Flash weights today?

Not yet. At the November 16, 2026 launch, Hy4 Flash is hosted-only across Tencent Cloud TokenHub, WorkBuddy, and OpenCode Zen; open-weight distribution has been signaled as a planned follow-up.

Information reflects public documentation as of November 17, 2026; verify live promotional deadlines, currency conversion, and model IDs in Tencent Cloud and OpenCode Zen before deploying.