iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Mingbird’s author reports that the open-source agent harness scored 0.821 with a 2B Gemma model on an 18-task benchmark, ahead of three other harnesses in that test. That is a promising result for locally run small models—not proof that any 2B model can reliably handle real work. The benchmark used one Windows machine and one trial per test cell, and its headline matrix ran on Mingbird v1.5.0, not the project’s currently listed v2.1.2.
What does Mingbird’s benchmark actually show?
The project’s 2026 report compares four agent harnesses across four open models and 18 tasks, producing 288 harness-model-task cells. With gemma4:e2b (2B), Mingbird scored 0.821 on the author’s LRAB benchmark; goose scored 0.271, opencode 0.017, and agent-mini 0.246. These are scores reported by Mingbird’s author, not independently verified findings. The benchmark article describes the experiment, and the project repository publishes the data and reproduction materials.
The comparison is useful as a specific question: under the author’s setup and scoring rules, how well did each harness turn model outputs into task deliverables? It does not establish that the same ranking holds for other tasks, computers, model configurations, or versions. The author describes the tier-level results as directional rather than statistical.
Recommended Free Tools
Author-reported LRAB scores
All values below are the project author’s 2026 results. They are benchmark scores, not percentages of tasks guaranteed to succeed.
#1 Best Overall
- EVOLUTION AMD RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 128GB pool, which is perfect for running LLMs such as Deepseek 70B Q8, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 12% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
| Model | Mingbird | goose | opencode | agent-mini |
|---|---|---|---|---|
| gemma4:e2b (2B) | 0.821 | 0.271 | 0.017 | 0.246 |
| qwen3.5:4b | 0.876 | 0.801 | 0.465 | 0.706 |
| gemma4:12b | 0.906 | 0.772 | 0.539 | 0.576 |
| ornith-1.5:35b | 0.941 | 0.679 | 0.896 | 0.092 |
| Overall | 0.886 | 0.631 | 0.479 | 0.405 |
The report also gives a separate small-model matrix: with qwen3.5:2b, Mingbird scored 0.779, compared with 0.239 for goose, 0.096 for opencode, and 0.017 for agent-mini. That is another author-run result on the project’s benchmark, not an independent replication.
How was the 288-cell test run?
The matrix covered four harnesses, four models, and 18 tasks. The author says runs used one Windows machine, a shared time budget, temperature 0, and thinking disabled. Scores were based on produced artifacts rather than transcripts. The benchmark code base is identified in the repository as tag v1.5.0, commit 1ee92d1; the repository provides the task set, scorer, raw per-cell data, and reproduction instructions.
The article reports that the computer used integrated graphics and 32 GB of RAM. That makes the result relevant to local inference on a conventional computer, but it is not a hardware-performance comparison: there is no evidence here about speed or success on other machines. The author also reports pinned backend details and collection windows in the repository; use those reproduction notes when matching the original campaign.
Free tools Windows power users keep installed
One-click scans. No signup required.
Version matters
The benchmark campaign used Mingbird v1.5.0. The repository’s README, observed October 7, 2026, lists v2.1.2 as the current version, while the original article referred to v1.8.2 as current when it was published. The published scores therefore should not be attributed to v2.1.2. The project’s releases page is the place to check current installer availability.
Rank #2
- Built for Local AI Development: AMD Ryzen AI Halo is designed for local AI development and inference, featuring 128GB unified memory and support for up to 200B parameter models to build and run intensive AI workloads locally.
- 128GB Unified Memory: Features 128GB LPDDR5x unified memory at 8000 MT/s with 256 GB/s memory bandwidth, providing a shared memory pool across the CPU, GPU, and NPU to support larger AI models.
- AMD Ryzen AI Max+ 395 Processor: Features 16 cores, 32 threads, and Zen 5 architecture, paired with AMD Radeon 8060S integrated graphics featuring 40 RDNA 3.5 compute units and an AMD XDNA 2 NPU with up to 50 TOPS.
- Linux AI Developer Platform: Purpose-built for Linux-based AI development with full AMD ROCm software support and preloaded tools, models, and workflows optimized for local AI development.
- Compact, Connected Design: Includes a 2TB M.2 SSD, 10GbE LAN, Wi-Fi 7, Bluetooth 5.4, USB-C connectivity, and HDMI 2.1b.
Why might a harness help a small model?
A model is only one part of an agent workflow. The harness routes tool calls, checks results, and decides whether work is finished. Mingbird’s author highlights three mechanisms: a finish gate that checks the original request against deliverables, verification feedback that returns real task-test errors, and anti-loop handling for repeated tool-call signatures or long turns without output. The article also describes a fixed 797-token factory prefill.
The intended benefit is to catch practical failure patterns—for example, a tool call that repeats without progress, or a model that says a task is complete while expected files are missing. These are the project’s stated design goals, not measured estimates of how often small models fail. The author says per-mechanism ablation differences were smaller than single-cell variance, so the benchmark does not establish how much any one mechanism contributes.
The README describes additional features such as automatic backups and rollback, bounded crawling of large files, chunked output feedback, and safeguards around destructive operations. These are project-documentation claims, not independently audited findings.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →What the scores do not prove
There are 18 tasks and one trial per cell, all on one machine. A single run cannot show how much scores vary between attempts, and the task set cannot represent every coding, research, file-management, or analysis workflow. The results also depend on the selected models, harness configurations, tools, backend, time budget, and artifact-based scoring criteria.
Rank #3
- EVOLUTION AMD RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 64GB pool, which is perfect for running LLMs such as Deepseek 32B, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 4% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
- Not a general guarantee: 0.821 does not mean a 2B model will complete 82.1% of your everyday jobs.
- Not a broad hardware result: one integrated-GPU computer does not show what to expect on every laptop or desktop.
- Not an independent evaluation: Mingbird’s author ran and reported the LRAB benchmark. The repository makes materials available for inspection, but public data alone does not make the original comparison independent.
- Not a current-version test: the headline matrix was run on v1.5.0, while the repository now lists v2.1.2.
The article also reports a WF-08 example in which a fix changed the result from 156 tool calls and zero files to 15 calls and three files, with identical bytes sent to the model. This is a single-task example, not an average improvement or a causal estimate across the benchmark. It separately reports author-run τ²-bench results of 0.763 for retail, 0.740 for airline, and 1.000 for telecom. Those are also project-run results, not independent evaluations, even though τ²-bench is identified as a Sierra Research benchmark.
How to decide whether the result matters for your work
The strongest test is whether the harness works on your own repeatable tasks. Before relying on a score, compare like with like: the model family and size, task type, deliverable criteria, tools and verification, inference backend, and hardware. A benchmark that rewards files or test outcomes may not predict performance on a task whose success is judged differently.
- Choose a small task with an objective result. For example, ask the agent to create a file with specified contents or make a bounded code change that can be checked by a test.
- Define completion before running it. List expected files, content, test results, or other observable deliverables so that a confident completion message is not mistaken for success.
- Repeat runs. Record failures as well as successes; one good run does not reveal consistency.
- Inspect the work and tool activity. Check that the deliverable is correct, commands were appropriate, and the agent did not make unrelated changes.
- Compare with another harness only under the same conditions. Keep the model, prompt, task, time budget, backend, and scoring criteria consistent, and note any configuration differences.
The project repository’s public benchmark materials can help readers inspect how the author defined tasks and scored artifacts. Your own results may differ because your tasks and setup may differ.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteWhat is Mingbird, and what does local-first mean here?
Mingbird describes itself as an open-source, local-first agent harness for Windows and Ollama, with a GUI and CLI and an Apache-2.0 license. The current README lists Windows 10 and 11 as the main platform, with Linux and macOS marked experimental. It documents Ollama and other OpenAI-compatible local servers. The project presents practical uses including writing and testing code, organizing files, web research, data analysis, and MCP tools.
Rank #4
- 【Leading AI Mini Workstation】MINISFORUM AI MS-S1 Max Workstation comes with AMD Ryzen AI Max+ 395 processor, which uses AMD's latest generation Zen 5 architecture. It has 16 Cores and 32 Threads, the boost clock is up to 5.1GHz. The overall processor performance is up to 126 TOPS, and the NPU performance reaches up to 50 TOPS. AMD Ryzen AI enables improved productivity, advanced collaboration, and improved efficiency.
- 【AMD Radeon 8060S Graphics 】The MS-S1 Max Mini PC equipped with AMD Radeon 8060S Graphics which built on the new generation of RDNA 3.5 architecture AMD graphics, it brings ultra-high frame rate experiences and advanced content creation features anywhere and delivers staggering performance. It can handle all your computing and multimedia tasks efficiently.
- 【Five 8K Video Output】This MS-S1 Max Workstation comes with five video outputs, 1x HDMI (8K@60Hz), 2x USB4(40Gbps,Alt DP2.0,PD out 15W) and 2x USB4 V2(80Gbps,Alt DP2.0,PD out 15W) Outputs, which support multiple monitors display at the same time and provide a larger and wider filed of view and improve your work efficiency. It is used in fields that require high-performance computing and graphics processing, including digital signage and securities trading, as well as work that uses CAD, such as engineering design, scientific calculations, animation production, and post-production for movies and television
- 【 Fast and Stable Wire & Wireless Speed】It comes with Two 10G Lan Ports for wired connection and and Wi-Fi 7 / BT5.4 for wireless connection, which increased the network speed greatly and expand its functions and improved performance of computer to a large extent and allows you to use more networks such as software routers (OpenWRT / DD-WRT / Tomato etc.), firewalls, NAT, network isolation etc.
- 【Large Storage & Flexible Expandability】This Workstation equipped with 64GB LPDDR5-8000MHz + 2TB M.2 2280 PCIe4.0 SSD. There is another PCIe4.0 SSD slot available for up to 8TB, these SSD slots are compatible with RAID0 and RAID1, you can store movies, videos, photos, important files easily. What’s more, it also comes with 1x standard PCIex16 slot(PCIe4.0x4) inside.
“Local-first” describes the default inference path, not an assurance that every feature stays offline. The project says it has no account requirement, telemetry, update check, or analytics, and documents an offline toggle that disables web tools and the optional cloud path. In normal online use, web search and user-configured MCP servers can send traffic outside the computer. These privacy and network-behavior statements come from the project documentation and have not been independently audited here.
The README warns that the agent reads and writes files and runs commands. Use a working directory whose contents you are willing to let the agent access, review proposed or completed changes, and keep backups for important data. Running locally does not remove the risk of mistaken edits or unsafe commands.
What computer do you need to try it?
The author’s run used a computer with integrated graphics and 32 GB RAM. The current README describes an entry tier of integrated graphics and 16 GB RAM for 2–4B models; that is the project’s stated guidance, not a guarantee of acceptable speed or a universal minimum. If you already have a suitable computer, try it before shopping. The benchmark does not establish that a new laptop is necessary or endorse any specific model.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
