
Hy4 flash chega a US$ 0,070 por milhão de tokens de entrada: o modelo Hy mais barato da Tencent também é o menos documentado
- GoogleNOVOGoogle: Gemini 4 Carbon2026-11-10$2.00 / $10.00 por 1M de tokens · 90 tok/s
- AnthropicNOVOAnthropic: Claude Fable 5.22026-11-09$8.00 / $40.00 por 1M de tokens · 97 tok/s
- orcaNOVOOrca: OrcaC2 1.02026-11-0717 tok/s
- OrcaNOVOOrcaCyber: GLM 5.3 Abliterated / Uncensored2026-11-04$3.00 / $5.00 por 1M de tokens · 87 tok/s
- OrcaNOVOOrcaCyber: Hy4 Abliterated / Uncensored2026-11-04$0.80 / $1.60 por 1M de tokens · 133 tok/s
- OrcaNOVOOrcaCyber: Kimi K3.1 Pro Abliterated / Uncensored2026-11-04$4.00 / $20.00 por 1M de tokens · 72 tok/s
- OrcaOrcaCyber: MiniMax M3.1 Flash Abliterated / Uncensored2026-10-29$0.80 / $1.60 por 1M de tokens · 130 tok/s
- OrcaOrcaCyber: GLM 5.3 Flash Abliterated / Uncensored2026-10-27$0.80 / $1.60 por 1M de tokens · 65 tok/s
- OrcaOrcaCyber: Qwen3.8 Flash Next Abliterated / Uncensored2026-10-26$0.80 / $1.60 por 1M de tokens · 92 tok/s
- OrcaOrcaCyber: MiMo V2.6 Flash Abliterated / Uncensored2026-10-26$0.80 / $1.60 por 1M de tokens · 43 tok/s
- MoonshotMoonshotAI: Kimi K3.1 Pro2026-10-2447Inteligência
- OrcaOrcaCyber: DeepSeek V4.1 Flash Abliterated / Uncensored2026-10-23$0.80 / $1.60 por 1M de tokens · 193 tok/s
- minimaxMiniMax: M3.1 Flash2026-10-1929Inteligência
- OrcaOrca: OrcaCyber Ultra 1.02026-10-17$5.00 / $10.00 por 1M de tokens · 150 tok/s
- deepseekDeepSeek: DeepSeek V4.1 Flash 10172026-10-1733Inteligência
- DeepSeekDeepSeek: DeepSeek V4.1 Flash2026-10-1740Inteligência
- OrcaOrca: OrcaCyber Zero 1.52026-10-10$3.00 / $7.50 por 1M de tokens · 119 tok/s
- openaiOpenAI: GPT-6.1 Sol2026-09-2955Inteligência
- anthropicAnthropic: Claude Sonnet 5.52026-09-2855Inteligência
- typesafeTypeSafe: Jev 1.132026-09-24$0.04 / $0.00 por 1M de tokens · 219 tok/s
Tencent released Hy4 flash on 16 November 2026, and the specification reads like a cost-engineering exercise more than a capability one: 224 billion total parameters of which 64 billion are Engram parameters, 12 billion active per token, native multimodality, a 1,048,576-token context, and a mainland list price of ¥0.5 per million input tokens against ¥2 per million output. The company's framing is that it matches Claude Haiku 5.5 on everyday work in a blind test run on its own employees, at roughly an eighteenth of the per-task cost. The blind test is real and the price is real; the comparison across them is Tencent's own arithmetic, and the difference between those two statements is most of what anyone deploying this model needs to know.
There is no model card. There is no architecture paper, no licence, no weights repository and no independent benchmark run. What Tencent published instead is a research page with benchmark results baked into PNG charts, a set of internal product testimonials, and a promise to open-source the model later. That is a thinner dossier than Hy4 preview shipped with in August, and the gap matters more here than it would for a flagship, because the entire pitch for Hy4 flash is efficiency — and efficiency claims are the ones you cannot check from a price table.
O que a tabela de tarifas realmente diz
Per million tokens, on the mainland card: ¥0.5 input, ¥2 output, ¥0.02 cached input. Tencent quotes the same model in dollars as $0.070 input, $0.281 output and $0.003 cached, which is the figure the launch post led with. Set it beside the rest of the family and the positioning is unambiguous.
• Hy4 flash — ¥0,5 de entrada / ¥2 de saída / ¥0,02 em cache, 224B no total com 12B ativos.
• Tencent HY4 — ¥6 input / ¥18 output / ¥0.3 cached, 770B total with 49B active, the general-availability build from 4 November 2026.
• Prévia do Hy4 — a mesma tarifa de ¥6 / ¥18 / ¥0,3, Apache 2.0, pesos disponíveis para download desde 28 de agosto de 2026.
Isso representa um corte de doze vezes na entrada e de nove vezes na saída em relação ao seu próprio carro-chefe, de uma geração para a outra, sem qualquer mudança na janela de contexto. A entrada em cache cai quinze vezes, que é o item de custo que decide a economia de qualquer agente que reenvia um longo prompt de sistema a cada etapa. Para uma carga de trabalho que lê um documento grande uma vez e depois faz vinte perguntas sobre ele, a taxa de cache é a taxa que importa, e US$ 0,003 por milhão está precificado para tornar esse padrão quase gratuito.
Two promotions sit on top of the list price and both are temporary. Tencent Cloud's TokenHub is running a 50% discount for a limited period with no stated end date, and WorkBuddy is free to use until 30 November. Neither should be modelled into a production budget; the list rate is the number that survives December.
De onde vêm 12 bilhões de parâmetros ativos

The parameter split is the first thing to understand, because 64 billion Engram parameters inside a 224-billion-parameter model is not a normal arrangement. Engram, in Tencent's usage, is a memory-style component that holds learned associations rather than participating in every forward pass the way an expert does. The practical consequence is that a large share of the model's total size is not charged against per-token compute: 12 billion parameters activate per token, which is a quarter of what GLM-5.3 Flash activates and roughly a quarter of Tencent HY4's 49 billion.
Tencent names four levers, and each one attacks a different part of the bill.
Pre-training. Hybrid attention combined with LatentMoE and the Engram parameters. Tencent's stated result is a KV cache one twenty-fourth the size of Hy3's and one sixth of Hy4's. On long-context serving that is the difference between a request fitting on a node and a request needing a cluster, and it is the single most consequential number in the release for anyone renting GPUs by the hour.
Post-training. Self-distillation from Tencent HY4, then reinforcement learning. The stated outcome is a 50% reduction in end-to-end output length. This is where a flash model usually gives back part of its price advantage — shorter answers can mean shallower work — and it is also where the per-task cost claims below get most of their leverage. Output tokens are the expensive ones at $0.281 per million, so halving them halves the larger side of the invoice before the price cut is even applied.
Inference. A megakernel, four-bit quantization and a decoding path Tencent calls DFly, together sustaining better than 150 tokens per second of decode. Fast decode changes agent behaviour as much as it changes user experience: a tool-calling loop that spends less wall-clock per step can afford more steps inside the same latency budget.
Detalhe da arquitetura. Nenhum publicado. A contagem de camadas, o número de especialistas, a largura do roteamento, o tamanho do codificador de visão e se a multimodalidade é nativa ou acoplada posteriormente estão todos ausentes. Os números de 224B e 12B vêm da página de pesquisa; tudo o que é mais granular não existe em público.
O teste cego, e o que "equivale ao Claude Haiku 5.5" abrange
The claim that travelled is the blind test. Tencent ran it with 85 users of its WorkBuddy product across 27 task categories and 259 real daily-work tasks, comparing Hy4 flash with several unnamed alternatives and reporting that user satisfaction came out close to Claude Haiku 5.5, that Hy4 flash had the lowest per-task cost of any model in the panel, and that it finished tasks in the shortest time.
Leia com atenção o texto, porque o post de lançamento e a página de pesquisa não dizem a mesma coisa. O post afirma que o modelo "acompanha" o Claude Haiku 5.5. A página diz que a satisfação ficou próxima disso. Próximo não é igual, e um fornecedor que resume seu próprio painel escolherá o enquadramento que o favorece. Mais importante ainda: o painel é composto por 85 funcionários da empresa que criou o modelo, trabalhando em tarefas que a empresa selecionou, em uma interface que a empresa também cria. Essa é uma forma razoável de decidir se um modelo deve ser lançado internamente. Não é evidência de que o resultado de terceiros será o mesmo, e não estabelece paridade em nenhum tipo de tarefa que as 27 categorias por acaso não incluíram.
O que o teste estabelece é mais restrito e ainda assim útil: em trabalho com formato de escritório — documentos, slides, planilhas, o tipo de coisa para a qual o WorkBuddy existe —, um modelo com 12B ativos não era obviamente pior do que um concorrente próximo da fronteira, a julgar por pessoas que usam os dois diariamente. Se isso se generalizar, o argumento de custo é esmagador. Se isso se generaliza é a questão em aberto, e não há nenhum benchmark público no lançamento que responderia a isso.
Rastreando o 18x
The "18x task cost reduction" in the launch post is not a price ratio. Hy4 flash costs twelve times less per input token than Tencent HY4, so a simple per-token comparison gives 12x, not 18x. The extra comes from Tencent's WorkIndex, an internal eight-suite benchmark covering PowerPoint creation, PowerPoint revision, human-machine co-writing, Markdown co-writing, document editing, table repair, audio and video processing, and visual understanding. On that suite, at the maximum tier and at busy-hour prices, Tencent reports per-task cost at one-thirteenth of Hy4 preview, one-twentieth of Tencent HY4, and one-eighth of DeepSeek-V4.1-Flash-1017.
One-twentieth against its own flagship and one-eighteenth against the model the post names is the same order of magnitude, so the 18x is a defensible round number — but it is a per-task figure, which means it folds in output length, retry rate and step count alongside price. Two of those three inputs are things Tencent controls through post-training and reports only as a percentage. A model that answers in half the tokens at a twelfth of the price is not 18x cheaper because of the price; it is 18x cheaper because it is also being made to talk less. If your task needs the longer answer, you get less than 18x, and the article you are reading cannot tell you how much less.
The same caveat applies to every per-task figure in the release, including the internal product numbers Tencent volunteered: a 70% cut in task time and roughly one-fifteenth the cost in WorkBuddy against Tencent HY4, a 2.52% general-chat gain against GLM-5.3-Flash in 元宝, 29% fewer output tokens and 6% fewer steps in 微信支付 agent flows against DeepSeek-V4.1-Flash-1017, 88% faster output and 60% fewer thinking tokens in 企业微信, and 22% to 39% fewer requests with 40% to 61% fewer tokens in 王者荣耀 root-cause analysis. These are the most credible numbers in the package, because they come from production systems with real users and real bills. They are also unaudited and selectively released, and no product publishes the flows where the model lost.
O que a versão não contém
No open weights. Tencent says it plans to open-source Hy4 flash and to iterate on it frequently. A plan is not a licence. Hy4 preview shipped under Apache 2.0 on day one with a Hugging Face repository, vLLM and SGLang recipes and published sampling defaults; Hy4 flash has none of that, and the wording around the open-source commitment is softer than the preview's was.
No independent evaluation. As of publication, Artificial Analysis has not posted a score for Hy4 flash, and no third party has reproduced a single number from the release. Every figure above is Tencent's.
Nenhuma tabela de benchmark legível por máquina. Os resultados do DeepSWE, Terminal-Bench 4.0, APEX-Agents, AutomationBench, HLE e OSWorld 2.1 existem como imagens de gráficos na página do fornecedor, não como texto. Essa é uma escolha estranha para um lançamento cujo argumento central é a eficiência mensurável, e significa que os números não podem ser copiados, comparados por diff ou verificados sem ler pixels.
Nenhum cartão de segurança ou do sistema. Também não há qualquer declaração sobre o comportamento de recusa, o tratamento de dados ou a disponibilidade regional para além dos produtos continentais mencionados.
Onde ele roda e como experimentá-lo a baixo custo
Availability on launch day is Tencent's own surface: TokenHub on Tencent Cloud for API access, WorkBuddy for office tasks free until 30 November, and 元宝, 企业微信, QQ, ima and Marvis on the consumer and enterprise side. Nothing else is announced.

On OrcaRouter, Hy4 flash is not in the catalogue yet — the route that exists for this family is tencent/hy4-preview, at $0.834 per million input and $2.501 per million output, alongside a rate-limited free tier. That is the model Hy4 flash was distilled from, so it is the right baseline to measure against while you wait: run the same tasks on both once the flash route lands and you will have the per-task ratio that Tencent's WorkIndex implies without having to trust it.
The reason to care about where it is served is the price model. OrcaRouter passes provider list price through at 0% markup, which is not a discount but a commitment about what happens when a vendor cuts one: Tencent has already run two promotional discounts on this model in its first week, and a pass-through route reflects the next one the day it appears rather than when a reseller updates a table. Failover matters for the same reason in reverse — an unproven 12B-active model on launch-week infrastructure is exactly the model you want to be able to fall back from without rewriting a client, and the same key reaches more than 200 models if it does.
Para onde apontá-lo primeiro?
Tencent's own deployment guidance splits the family cleanly, which is more useful than most vendor advice because it concedes a weakness: Hy4 flash for deep search, document processing, PowerPoint generation, simple frontend work and computer use; Tencent HY4 for long-horizon software engineering. The concession is in the last clause — the cheap model is not being sold as a coding replacement for the expensive one, and the internal WorkBuddy data shows a small drop on long coding tasks even as it reaches parity on daily work.

Isso se encaixa em um plano de teste simples. Pegue vinte tarefas que você já executa, das quais a maioria é relacionada a documentos ou típica de agentes e algumas são genuinamente difíceis de engenharia. Registre tokens de entrada, tokens de saída, acertos de cache, tempo total decorrido e se a saída foi utilizável. Execute-as no Hy4 flash e no baseline que você usa hoje. Em uma tarde, você vai descobrir se o 18x se aplica ao seu tráfego, o que é mais do que o próprio lançamento vai lhe dizer.
O modelo Hy mais barato é o menos legível.
Tencent has built something worth paying attention to: a quarter-active model with a million-token window, a KV cache small enough to change the serving maths, and a price that undercuts its own flagship by an order of magnitude. The engineering described — Engram parameters, a 50% output-length cut through self-distillation, four-bit inference at over 150 tokens per second — is the kind of work that shows up in a bill rather than a leaderboard, which is presumably why Tencent led with per-task cost.
But the release asks to be believed on exactly the quantities it declines to document. No weights, no model card, no independent score, benchmark numbers behind pixels, and a headline comparison drawn from a panel of its own employees. None of that is dishonest. All of it means the honest summary is "this is probably the best value in the Hy family and nobody outside Tencent has verified it," and the right response is a two-week trial with token counts, not a migration.
The open-source promise is the thing to diary. If Tencent ships weights the way it did for Hy4 preview — Apache 2.0, day one, with serving recipes — the efficiency claims become checkable and the model becomes a serious option for anyone who wants to run cheap inference themselves. Until that happens, Hy4 flash is a very good price attached to a very thin dossier, and the difference between those two facts is where your evaluation budget should go.
Comparados neste artigo3
Detectado a partir deste artigo · Benchmarks: Artificial Analysis · atualizado diariamente
