Lichess vs. Stockfish: Why Online Analysis Shows Higher Speed but Slower Depth

Lichess’s Stockfish engine reports a higher node‑per‑second (N/s) rate than a native Stockfish binary, yet it takes longer to reach the same search depth because it runs in a WebAssembly (WASM) environment, often uses Multi‑PV mode, and employs a smaller evaluation net.

On Hacker News a lively discussion erupted around a puzzling performance gap: the browser‑based analysis board on Lichess reports close to 1 MN/s on a Redmi Note 14 Pro, while a locally‑run Stockfish binary via Python only shows about 600 kN/s. Paradoxically, Lichess needs roughly 2 minutes 30 seconds to hit depth 30, whereas the local setup reaches the same depth in under a minute. This article unpacks the technical reasons, community insights, and practical takeaways for developers and chess enthusiasts.

Key Takeaways from the Thread

• Multi‑PV mode: Lichess often runs Stockfish with Multi‑Principal Variation (default 5 lines), which spreads the search across several branches, inflating N/s but reducing depth per line.
• Evaluation net size: The default Lichess net is a trimmed version of Stockfish’s neural network, designed for speed on mobile browsers.
• WASM vs. native binary: WebAssembly introduces a small overhead and limits thread usage compared with a compiled native executable.
• Thread count & hash tables: Lichess caps threads (often 2‑4 on mobile) and uses modest hash sizes, whereas a desktop Python script may allocate more cores and larger transposition tables.
• UI‑driven reporting: The N/s figure shown on Lichess is an instantaneous snapshot, not an average over the whole analysis session.

Technical Deep‑Dive: What Drives the Numbers?

1. Multi‑PV and Search Breadth

When Multi‑PV is enabled, Stockfish evaluates several top moves simultaneously. Each extra line multiplies the node count, so the engine reports a higher N/s. However, the search tree is divided, meaning each line progresses more slowly toward deeper depths. As ChatGPT and Telegram integration experts often note, breadth can mask depth in performance metrics.

2. Neural‑Network (NNUE) Size

Lichess ships a compact NNUE model to keep download sizes low for browsers. The smaller net evaluates positions faster but with slightly less precision, which can affect the engine’s pruning efficiency. A larger net, selectable in the Lichess UI, yields lower N/s but deeper, more accurate analysis—mirroring the local binary’s behavior.

3. WebAssembly Overhead

WASM runs inside the browser’s sandbox, translating machine code on the fly. While modern browsers have optimized this path, there remains a measurable latency compared to a native executable compiled for the host CPU. Additionally, browsers limit the number of Web Workers, capping parallelism. This is why a native OpenAI ChatGPT integration can fully exploit all CPU cores.

4. Hash Table & Transposition Table Settings

Stockfish’s hash size directly influences how many previously evaluated positions it can store. Lichess uses a modest default (≈ 16 MiB) to conserve mobile memory, while a desktop script may allocate 256 MiB or more, dramatically reducing re‑search of identical positions and speeding up depth growth.

5. Reporting Mechanics

The “N/s” shown on Lichess updates every second, reflecting the current throughput. In contrast, many local GUIs display an average over the entire session. This discrepancy can mislead users into thinking the engine is faster when, in reality, the average speed may be lower.

Understanding these nuances is essential for anyone building AI‑enhanced chess tools, whether you’re integrating Stockfish into a SaaS product or creating a personal analysis pipeline.

What This Means for Chess Enthusiasts and Developers

For casual players, the higher N/s on Lichess is mostly cosmetic; the real metric of interest is the depth and quality of the suggested moves. Developers, however, must consider the following:

1. Choose the right mode: If you need deep analysis, disable Multi‑PV and select the “large” NNUE net in Lichess or configure your local engine accordingly.
2. Allocate resources wisely: On server‑side deployments, increase thread count and hash size to match the hardware capabilities. The Workflow automation studio can orchestrate such resource tuning automatically.
3. Consider deployment environment: WASM is ideal for cross‑platform web apps, but for heavy‑duty analysis (e.g., tournament preparation) a native binary or a cloud‑based container is preferable.
4. Leverage AI‑enhanced pipelines: Combining Stockfish with language models (e.g., via AI marketing agents) can generate natural‑language explanations of engine evaluations, adding value for end‑users.

Highlighted Comments from the Discussion

“Likely what’s going on is that your Lichess instance is running in MultiPV mode, which displays the top N moves (5 by default). This can be useful for analysis, but it leads to more time exploring disfavored lines, and therefore lower depth.” – anematode (Stockfish contributor)

“The reason is because the SF17.1 is functionally different, there is no way to get these to match without changing the net option to be the same as the Stockfish binary.” – Viren6 (Stockfish contributor)

“Lichess uses a smaller net by default. Default lichess net is custom made by Stockfish to be small.” – Viren6

How UBOS Can Help You Build Smarter Chess Tools

Developers looking to create custom analysis platforms can benefit from the UBOS platform overview, which offers a modular architecture for integrating engines like Stockfish. The platform’s Web app editor on UBOS lets you spin up a browser‑based analysis board with just a few clicks, while the Enterprise AI platform by UBOS provides scaling options for high‑throughput cloud deployments.

If you need to add voice commentary to your analysis, the ElevenLabs AI voice integration can synthesize natural‑sounding explanations of engine lines. For data‑intensive workloads, the Chroma DB integration offers fast vector search for storing and retrieving millions of positions.

Startups can accelerate their go‑to‑market with UBOS for startups, while SMBs benefit from pre‑configured solutions via UBOS solutions for SMBs. Pricing is transparent on the UBOS pricing plans page, and you can explore real‑world use cases in the UBOS portfolio examples.

For rapid prototyping, the UBOS templates for quick start include a ready‑made Stockfish integration template that handles thread allocation, hash configuration, and optional Multi‑PV toggling. Pair it with the Telegram integration on UBOS to deliver analysis results directly to a chat bot, or combine with the ChatGPT and Telegram integration for natural‑language explanations.

Conclusion: Choose the Right Tool for the Right Job

The apparent speed advantage of Lichess’s Stockfish is a product of its Multi‑PV mode, smaller neural net, and the constraints of a WASM environment. For developers who need raw depth and consistent performance, a native Stockfish binary with tuned hash tables and full‑core utilization remains the gold standard. However, for web‑centric products that prioritize accessibility and instant feedback, Lichess’s setup offers a compelling user experience.

Whether you’re building a personal analysis suite or a commercial AI‑driven chess service, consider the trade‑offs outlined above and leverage platforms like UBOS homepage to accelerate development.

Ready to prototype your own AI‑enhanced chess analyzer? Explore the UBOS partner program today and turn these insights into a market‑ready solution.