{"slug": "gamer-vla-model-mihoyo-s-long-horizon-game-agent-explained", "title": "Gamer VLA Model: miHoYo's Long-Horizon Game Agent Explained", "summary": "MiHoYo's Honkai AI R&D team released Gamer, a vision-language-action foundation model that plays long 3D games from screen pixels using keyboard and mouse commands without a hidden game-state API, publishing the arXiv paper on November 13, 2026 alongside Apache-2.0 code on GitHub. Gamer was trained through continued pre-training on tens of thousands of hours of gameplay, sparse-reasoning supervised fine-tuning, and Temporal-Aware GRPO reinforcement learning, served via Streaming-vLLM with KV reuse and speculative decoding that the README cites as a roughly 20× latency drop on H200 hardware with Qwen3.5-35B-A3B MoE, targeting sub-100 ms decision loops. The work reached #2 on Hugging Face Daily Papers, with the team arguing the same action vocabulary should work across 2D benchmarks and 3D open worlds and that trajectories can mix virtual games with embodied simulators.", "body_md": "**Gamer is miHoYo's new [vision-language-action (VLA)](https://explainx.ai/dictionary/vision-language-action-model-vla) foundation model that plays long 3D games from screen pixels and human-style keyboard and mouse commands — no hidden game-state API.** The Honkai AI R&D team posted the [arXiv paper](https://arxiv.org/abs/2611.18397?ref=explainx) on **November 13, 2026**, opened the [project page](https://mihoyo-gamer.github.io/?ref=explainx), and shipped [Apache-2.0 code](https://github.com/RainBowLuoCS/Gamer?ref=explainx) that documents how a single policy reasons, remembers, and acts across **hour-long quests**.\n\nFor builders, Gamer is not a miHoYo storefront feature. It is a **research stack** — continued pre-training on **tens of thousands of hours** of gameplay, **sparse-thinking** supervision, **Temporal-Aware GRPO (TA-GRPO)** reinforcement learning, and **Streaming-vLLM** serving aimed at **sub-100 ms** decision loops. That combination is why the Hugging Face Daily Papers entry hit **#2** with rising upvotes: it connects [embodied agent roadmaps](https://explainx.ai/blog/envisioning-agentic-robotics-survey-s0-s1-s2-roadmap-2026) to something millions of players already understand — **beating a hard RPG quest under time pressure**.\n\n## TL;DR\n\n| Question | Answer | \n|---|---|\n| Who released it? | **miHoYo Honkai AI R&D Team** (authors on arXiv and GitHub) | \n| What interface? | **Screen observations** in,**keyboard and mouse** action tokens out | \n| Training recipe? | **CPT** on massive trajectories →**sparse-reasoning SFT** →**TA-GRPO** online RL | \n| Serving? | **Streaming-vLLM** with KV reuse and speculative decoding; README cites**~20×** latency drop on H200/Qwen3.5-35B-A3B MoE | \n| Code? | [GitHub RainBowLuoCS/Gamer](https://github.com/RainBowLuoCS/Gamer?ref=explainx) — prompts, parsers, benchmark tables; weights not bundled as a product drop | \n| How does this compare to cheat headlines? | Unlike [GPT-6 Astra's StarCraft shortcut](https://explainx.ai/blog/gpt-6-astra-starcraft-starskirmish-downloads-human-bot-cheat-2026) , Gamer's setup forbids swapping in human bots mid-episode; rewards come from**environment verification** | \n\n## Why miHoYo built a game-native VLA now\n\nGames are **stress tests** for agents: partial observability, **long horizons**, sparse rewards, and **real-time control** at 30–60 FPS. Productivity [computer-use agents](https://explainx.ai/blog/alibaba-opengcua-open-stack-generalist-computer-use-agents-osworld-v2-2026) often click through forms. RPGs demand **memory** (“which door did I already try?”), **combat sequencing**, and **navigation** across maps that do not fit one context window.\n\nmiHoYo's angle in the [abstract](https://arxiv.org/abs/2611.18397?ref=explainx) is **generalist control**: the **same action vocabulary** (mouse deltas, key holds, scroll) should work in **2D benchmarks** and **3D open worlds**, and trajectories can mix **virtual games** with **embodied simulators** so dynamics transfer. That mirrors how [Microsoft's idle-GPU gaming patent](https://explainx.ai/blog/microsoft-patent-idle-gaming-gpu-micro-training-us-20260349820-2026) and [Xbox Labs' AI porting experiments](https://explainx.ai/blog/xbox-labs-ai-game-decompiling-recompiling-ports-asha-sharma-2026) treat commercial titles as **data factories** — except Gamer publishes the **training and inference recipe**, not just IP around it.\n\n**What games did miHoYo make?** The studio is known for **Genshin Impact**, **Honkai: Star Rail**, **Zenless Zone Zero**, and related live-service titles. **Gamer is not a new game in that list**; it is an **R&D agent** evaluated partly inside **Honkai: Star Rail** quests, with demo art on the project page referencing multiple franchises for **benchmark variety**.\n\n## Architecture: perception, sparse reasoning, and memory\n\nThe policy is end-to-end: **vision encoder + language backbone + action head** that emits structured keyboard and mouse chunks. Three ideas separate it from a vanilla VLM fine-tune:\n\n1. **Unified visual control** — actions are**device-level** trajectories aligned to frames, not abstract joystick vectors tied to one title.\n2. **Sparse reasoning** — the model can output**action-only** tokens when the prior plan still works, and only opens a reasoning block when objectives change (documented in the[GitHub system prompt template](https://github.com/RainBowLuoCS/Gamer?ref=explainx) ).\n3. **Persistent memory summaries** — when the context window fills, the agent writes**compact memory** of finished subgoals, failures, and remaining plan steps, then continues rolling forward.\n\n## Sparse reasoning and cross-context memory\n\nGamer VLA model sparse reasoning and cross-context memory architecture diagram\n*Image: [miHoYo Honkai AI R&D Team](https://mihoyo-gamer.github.io/?ref=explainx)*\n\nThe figure above (from the official project page) shows how **reasoning** and **memory updates** share the **same autoregressive stream** as low-level actions. That matters for deployment: you do not run a separate planner service that drifts from the executor.\n\nPractically, each decision covers about **200 ms** split into **six ~33 ms steps**. Mouse movement applies to the whole chunk; keys can differ per step. The README lists **74 case-sensitive key tokens** and strict parsing rules — including treating malformed outputs as **no-ops**, which is how you keep an RL farm from melting down when the policy hallucinates format.\n\nIf you build agents, steal the pattern: **one tokenizer for thought, memory, and act**, with the harness enforcing **FSM-shaped structure** so speculative decoding can propose safe delimiters.\n\n## Training stages: CPT, SFT, and TA-GRPO\n\nThe [GitHub method table](https://github.com/RainBowLuoCS/Gamer?ref=explainx) breaks training into three supervised regimes plus online RL:\n\n| Stage | Supervision | What the policy learns | \n|---|---|---|\n| **Continued pre-training (CPT)** | Screen frames paired with keyboard and mouse from diverse games plus general multimodal data | Transferable visual control and dynamics before task-specific rewards | \n| **Sparse-reasoning SFT** | Multi-level instructions with **~1.2M reasoning** and**~223K memory** annotations (per Hugging Face team post) | When to think, what to remember, and how to finish subgoals | \n| **TA-GRPO** | Online rollouts with **verifiable environment outcomes** | Long-horizon credit assignment across segments linked by memory | \n\n**Temporal-Aware GRPO** is the paper's answer to **context limits** and **rollout error accumulation**: instead of pretending an hour-long episode fits one forward pass, training stitches **memory-linked segments** and assigns reward along the timeline. That is the game analogue of [multi-turn agent RL](https://explainx.ai/blog/uni-robot-agent-open-qwen-robodojo) harnesses in robotics — except rewards are **quest milestones** and **minigame win conditions** you can check automatically.\n\nThe team also continues a parallel **GameScaling** CPT line ([arXiv:2610.28698](https://arxiv.org/abs/2610.28698?ref=explainx), cited in the repo) on **scaling laws in diverse virtual worlds**, which explains why they emphasize **10K+ hours** of trajectory CPT in public messaging.\n\nGamer VLA multi-stage gameplay data curation and annotation pipeline\n*Image: [miHoYo Honkai AI R&D Team](https://mihoyo-gamer.github.io/?ref=explainx)*\n\n## Honkai: Star Rail results the README publishes\n\nGamer foundation model Honkai Star Rail side quest benchmark results chart\n*Image: [miHoYo Honkai AI R&D Team](https://mihoyo-gamer.github.io/?ref=explainx)*\n\nOn held-out **Honkai: Star Rail** tasks, the README posts concrete **milestone completion rates** after each stage. Side quests **Opportunity Makes the Thief (Part I)** and **Bestial Ferocity** climb from **34.4% / 47.9%** after CPT to **80.0% / 83.6%** after RL. **Minigame macro-averages** rise from **15.0%** to **49.7%**.\n\nThose numbers are **internal eval tables**, not a promise that a retail copy of Star Rail will ship a autopilot mode. They do show **TA-GRPO adds double-digit points** beyond SFT — the kind of lift you want before trusting an agent on **55-milestone** quests.\n\nThe Hugging Face announcement further claims **over 95%** success on **main-storyline challenges** in large-scale automated tests. Treat that as **author-reported coverage** until independent replications appear; still, it signals the model is meant for **production-scale QA automation**, not couch play.\n\n## Streaming-vLLM: why inference speed becomes a training variable\n\nLong-horizon RL fails if **action latency** blows the game's tick budget. Gamer's serving stack, **Streaming-vLLM**, targets **concurrent online rollouts** with:\n\n- **Per-environment KV-cache reuse** and memory compaction\n- **Verified finite-state-machine speculative decoding** for structural tokens\n- **Early action dispatch** so the client executes the first valid chunk while later tokens stream\n- **Architecture-specific paths** for**Qwen3-VL** and**Qwen3.5** dense and MoE backbones\n\nThe README's headline table compares **1306.00 ms** baseline latency to **65.12 ms** after cumulative optimizations on **NVIDIA H200** with **Qwen3.5-35B-A3B MoE** and **TP4** — about **20.1×** faster. The Hugging Face comment rounds this to **20×** and argues faster inference **improves online RL**, not just player feel.\n\nImplementation is marked **coming soon** in the `streaming-vllm/` directory. Until it lands, treat the engine as a **paper plus design doc**, not something you can npm install for Friday's demo.\n\n## How to try Gamer as a builder\n\n1. **Read the paper** —[arXiv:2611.18397](https://arxiv.org/abs/2611.18397?ref=explainx) for TA-GRPO and data-curation figures.\n2. **Clone the repo** —[github.com/RainBowLuoCS/Gamer](https://github.com/RainBowLuoCS/Gamer?ref=explainx) for system prompts, action grammar, and benchmark tables.\n3. **Watch the project page demos** —[mihoyo-gamer.github.io](https://mihoyo-gamer.github.io/?ref=explainx) hosts gameplay films (third-party game assets remain their owners' IP; Apache-2.0 covers code, not art).\n4. **Reproduce parsing, not full quests** — start by implementing the**action parser** and**sparse reasoning gating** on a toy 2D env before attaching a 35B MoE.\n\nThere is **no Hugging Face model card with open weights** attached to the November drop. The repo's roadmap still tracks **Streaming-vLLM code**, extra demos, and documentation. Plan for **research access**, not a Battle.net-style bot.\n\n### Action format cheat sheet\n\nEach chunk looks like:\n\n`<|action_start|>X Y Z ; g1 ; g2 ; g3 ; g4 ; g5 ; g6<|action_end|>`\n\n**X** and **Y** are mouse deltas in thousandths of screen width and height (clipped to ±1000). **Z** is scroll. **g1…g6** are key groups per 33 ms step. Reasoning, when needed, lives in a `<think>` block immediately before the action block — the server must parse **only the final block after the last closing tag** so it does not execute examples nested inside thoughts ([README parsing section](https://github.com/RainBowLuoCS/Gamer?ref=explainx)).\n\n## Benchmarks beyond one RPG\n\nProject-page art references evaluations across **Genshin Impact**, **Zenless Zone Zero**, **Minecraft**, **DOOM Eternal**, and classic **2D** suites — a spread designed to show **interface generalization**, not overfitting one UI skin. The paper abstract claims **competitive 2D benchmark scores** while emphasizing **3D open-world generalization**.\n\nThat positioning sits next to [WorldFoundry's multi-benchmark orchestration](https://explainx.ai/blog/worldfoundry-open-world-model-benchmark-infrastructure-2026) trend: games become **standardized agent gyms** when you can verify rewards programmatically.\n\n## Safety, cheating, and live-service risk\n\nGamer's research story lands next to Korea's **generative-AI exam cheating** scandal and everyday **anti-cheat** arms races. Technically, a screen-and-mouse policy is **indistinguishable from a skilled human** at the input layer — which is why online games rely on **server authority**, **behavioral stats**, and **kernel anti-cheat**, not “please don't use AI.”\n\nFor builders shipping agents into **live games**:\n\n- Assume **Terms of Service** ban unsanctioned automation even when inference is local.\n- Prefer **offline simulators** or**first-party QA environments** with signed clients.\n- Log **provenance** of trajectories if you fine-tune further — publishers increasingly audit**synthetic play data** .\n\nThis is parallel to why explainx.ai covered [Astra's StarCraft rule break](https://explainx.ai/blog/gpt-6-astra-starcraft-starskirmish-downloads-human-bot-cheat-2026) as a **harness failure**, not a skill flex.\n\n## What we still do not know\n\n- **Weight release scope** — paper and code are public;**checkpoint hosting** is not guaranteed.\n- **Streaming-vLLM open-source date** — README says**coming soon** .\n- **Independent replication** of**95%+ storyline** claims on consumer hardware.\n- **Latency numbers** outside the quoted**H200 / Qwen3.5-35B-A3B MoE** configuration.\n\nWatch the [GitHub releases](https://github.com/RainBowLuoCS/Gamer?ref=explainx) and the [Hugging Face paper thread](https://huggingface.co/papers/2611.18397?ref=explainx) for weights, not rumor accounts.\n\n## Related reading\n\nReading gets you the idea. Building it with feedback gets you the skill.", "url": "https://wpnews.pro/news/gamer-vla-model-mihoyo-s-long-horizon-game-agent-explained", "canonical_source": "https://explainx.ai/blog/mihoyo-gamer-vla-foundation-model-long-horizon-game-agent-2026?utm_source=rss&utm_medium=feed&utm_campaign=blog", "published_at": "2026-11-16 00:00:00+00:00", "updated_at": "2026-11-16 08:22:16.650882+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "ai-research", "large-language-models", "ai-infrastructure"], "entities": ["miHoYo", "Honkai AI R&D Team", "Gamer", "arXiv", "GitHub", "Hugging Face", "Qwen3.5-35B-A3B", "Honkai: Star Rail"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/gamer-vla-model-mihoyo-s-long-horizon-game-agent-explained", "markdown": "https://wpnews.pro/news/gamer-vla-model-mihoyo-s-long-horizon-game-agent-explained.md", "text": "https://wpnews.pro/news/gamer-vla-model-mihoyo-s-long-horizon-game-agent-explained.txt", "jsonld": "https://wpnews.pro/news/gamer-vla-model-mihoyo-s-long-horizon-game-agent-explained.jsonld"}}