{"slug": "is-workbuddy-free-how-to-try-hi-4-preview-at-no-cost", "title": "Is WorkBuddy Free? How to Try Hi-4 Preview at No Cost", "summary": "WorkBuddy, a desktop AI agent app, is offering free access to Tencent's Hi-4 Preview model for two weeks from launch, allowing users to try the 770 billion parameter mixture-of-experts model without payment. The model, released under Apache 2.0 with weights on Hugging Face, has API pricing of $0.834 per million input tokens and $2.001 per million output tokens, with cache hits at $0.042 per million tokens, making it one of the cheapest flagship openweight models. Local hosting is impractical for most due to the 1.8 TB BF16 weights or 900 GB FP8 memory requirement.", "body_md": "# Is WorkBuddy Free? How to Try Hi-4 Preview at No Cost\n\nWorkBuddy offers Tencent's Hi-4 Preview model free for two weeks. Here's how the trial works and what the API costs elsewhere.\n\n## Is WorkBuddy free right now?\n\nYes, for a limited window. WorkBuddy, a desktop AI agent app for office work, is giving users free access to Tencent’s newly released Hi-4 Preview model for two weeks starting from the model’s launch. You download the app, register, pick Hi-4 Preview from the model selector in the chat box, and start using it without paying. After the trial window closes, using Hi-4 through WorkBuddy or any other channel means paying per-token API rates, which are notably cheap compared to other flagship openweight models.\n\n## TL;DR\n\n- **WorkBuddy** is offering free access to Tencent’s Hi-4 Preview model for two weeks from launch, no payment required during that window.\n- **Hi-4 Preview** is a 770 billion parameter mixture-of-experts model with 49 billion active parameters per token, released under Apache 2.0 with weights on Hugging Face.\n- **API pricing** runs $0.834 per million input tokens and $2.001 per million output tokens, with cache hits priced at just $0.042 per million tokens, making it one of the cheapest flagship openweight models available.\n- **Local hosting is impractical for most people** since the BF16 weights are around 1.8 TB and even the quantized FP8 build needs roughly 900 GB of memory, meaning multiple high-end GPUs at minimum.\n- **Alternative access points** include Tencent Cloud’s Token Hub and OpenRouter, both of which let you plug Hi-4 into coding agents like Claude Code or other tools.\n- **WorkBuddy itself functions as an agent app** , breaking tasks into subtasks, running parallel agents, using a browser, reading local files, and producing finished output files like reports and spreadsheets.\n\n## Other agents ship a demo. Remy ships an app.\n\nReal backend. Real database. Real auth. Real plumbing. Remy has it all.\n\n## What exactly is Hi-4 Preview?\n\nHi-4 Preview is Tencent’s next-generation open-weight model, built as a mixture of experts with 770 billion total parameters but only 49 billion active per token. It uses 256 routed experts plus one shared expert, activating eight experts at a time. A gated sparse attention mechanism lets it handle a context window past 1 million tokens without inference costs spiraling. The model ships under the Apache 2.0 license with weights available on Hugging Face in both BF16 and FP8 formats.\n\nUnlike a general chat model, Hi-4 Preview is positioned specifically for agentic work: coding agents, multi-step planning, long-horizon task execution, tool calling, and office productivity tasks like building spreadsheets or slide decks. Tencent has also claimed the model helped optimize its own training pipeline, including kernel-level improvements that increased end-to-end throughput by around 31.8% against their baseline.\n\n## How does the pricing compare to other openweight models?\n\nThe API pricing is where Hi-4 stands out. At $0.834 per million input tokens and $2.001 per million output tokens, with cache hits at $0.042 per million tokens, it undercuts comparable openweight flagships like Qwen 3.8 Max and Kimi K3. DeepSeek V4 Flash remains cheaper if the goal is the absolute lowest cost option, but among models competing at the top of the openweight benchmark tables, Hi-4 currently offers the best price-to-performance ratio.\n\nThis pricing matters because of the Apache 2.0 license. Providers can host the model cheaply and pass savings to users, which is a large part of why it’s showing up in multiple access points (Tencent Cloud’s Token Hub, OpenRouter, and now WorkBuddy) so quickly after release.\n\n## Can you run Hi-4 Preview locally?\n\nRealistically, no, not on consumer hardware. The BF16 weights weigh in around 1.8 TB, and even the quantized FP8 build needs roughly 900 GB of memory. That puts local deployment in the territory of multiple high-end data center GPUs (the kind of setup that costs more than most people will ever spend on a home lab). For nearly everyone, accessing Hi-4 means going through an API or an app that hosts it for you, which is exactly why the WorkBuddy free trial and the OpenRouter/Tencent Cloud options matter for individual users and small teams.\n\n## How does Hi-4 Preview perform on benchmarks?\n\n- ✕a coding agent\n- ✕no-code\n- ✕vibe coding\n- ✕a faster Cursor\n\nThe one that tells the coding agents what to build.\n\nHi-4 Preview sits near the top of the current openweight pack. It scores 92.3 on GPQA Diamond, 85.44 on Terminal-Bench, and 82.9 on SWE-Bench Multilingual. It also leads the openweight field on some newer benchmarks, including Horizon Math and BioMystery Bench. Tencent ran an internal blind evaluation with 163 experts across 203 real engineering tasks, where Hi-4 Preview scored 2.99 out of 4, compared to 2.94 for Kimi K3 and 2.92 for GLM 5.3. That internal eval should be read with some caution since Tencent designed and ran it, but the public benchmark numbers suggest the model trades wins with Kimi K3 and GLM 5.3 while remaining behind closed models like Opus 5 and GPT 5.6 on the hardest tasks, which is typical for openweight models competing against frontier closed ones.\n\n## What is WorkBuddy and how does it use Hi-4?\n\nWorkBuddy is a desktop app built around AI agents for office work, comparable in concept to other “agent-for-work” tools. It takes a task, breaks it into subtasks, and spins up multiple agents that run in parallel, using a browser, reading local files, running code, and delivering finished output like reports, slide decks, spreadsheets, and small web apps. It supports enableable “skills” for research, coding, and analytics, and connects to services like GitHub, Notion, Google Drive, Gmail, and Slack.\n\nTo use Hi-4 Preview inside it, you download the app for Mac or Windows, register, log in, and choose Hi-4 Preview from the model selector at the bottom of the chat box. Switching models later, to compare outputs, is a matter of changing that same dropdown without altering anything else about the setup.\n\n## Is Hi-4 Preview worth trying during the free window?\n\nFor anyone doing agentic work like coding tasks, long document audits, or multi-source research summarized into slides, the free two-week window is a low-risk way to test a model that’s currently priced well below most comparable openweight competitors and benchmarks near the top of that category. Given that local hosting is out of reach for most individuals due to the memory requirements, this kind of temporary free access through an app is one of the more practical ways to evaluate the model’s real-world agentic performance before deciding whether ongoing API costs make sense for a particular workflow.\n\n## Frequently Asked Questions\n\n### How long is Hi-4 Preview free on WorkBuddy?\n\nThe free access window runs for two weeks starting from the model’s launch date. After that, standard API pricing applies.\n\n### What does Hi-4 Preview cost after the free trial?\n\nThe API pricing is $0.834 per million input tokens, $2.001 per million output tokens, and $0.042 per million tokens for cache hits.\n\n### Can I use Hi-4 Preview outside of WorkBuddy?\n\nYes. It’s available through Tencent Cloud’s Token Hub and through OpenRouter, which lets you connect it to coding agents and other tools.\n\n### Do I need special hardware to run Hi-4 Preview myself?\n\nEffectively yes. The BF16 weights are about 1.8 TB and the quantized FP8 version needs around 900 GB of memory, requiring high-end data center GPU setups rather than consumer hardware.\n\n### How does Hi-4 Preview compare to closed models like GPT or Opus?\n\nIt trades wins with other top openweight models like Kimi K3 and GLM 5.3 on public benchmarks, but it still trails closed frontier models like Opus 5 and GPT 5.6 on the hardest tasks.", "url": "https://wpnews.pro/news/is-workbuddy-free-how-to-try-hi-4-preview-at-no-cost", "canonical_source": "https://www.mindstudio.ai/blog/workbuddy-ai-free-access/", "published_at": "2026-09-08 00:00:00+00:00", "updated_at": "2026-09-08 20:53:38.519917+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-products", "ai-tools", "ai-infrastructure"], "entities": ["WorkBuddy", "Tencent", "Hi-4 Preview", "Hugging Face", "Tencent Cloud", "OpenRouter", "Qwen 3.8 Max", "Kimi K3"], "alternates": {"html": "https://wpnews.pro/news/is-workbuddy-free-how-to-try-hi-4-preview-at-no-cost", "markdown": "https://wpnews.pro/news/is-workbuddy-free-how-to-try-hi-4-preview-at-no-cost.md", "text": "https://wpnews.pro/news/is-workbuddy-free-how-to-try-hi-4-preview-at-no-cost.txt", "jsonld": "https://wpnews.pro/news/is-workbuddy-free-how-to-try-hi-4-preview-at-no-cost.jsonld"}}