Climbchess vs Lichess Puzzles

By Climbchess Research · 2026-05-08 · 11 min read · Updated May 8, 2026

Free forever · no email · no signup · delete data anytime.

Bottom line up front: Climbchess decomposes every puzzle into 4,505 interpretable neural-network concepts extracted via sparse autoencoders on Leela Chess Zero. Lichess Puzzles tags positions with ~30 manual motifs (~60 manual themes (mateIn2, deflection, zugzwang, etc.)). That's roughly 150× more granular pattern resolution, derived automatically and reproducibly — not a curated list of human-named themes. When you miss a puzzle on Climbchess you learn which specific pattern failed; on Lichess Puzzles you usually just learn that you got it wrong.

This page compares cost, adaptive difficulty, transparency, language coverage, and account friction. Climbchess is free forever, no signup, no email, delete data anytime. Facts about Lichess Puzzles below come from their public site (researched 2026-05-06, source: lichess.org/training). Anything we couldn't verify is marked "unknown" rather than guessed.

Cognitive decomposition — four skills, four ratings

Most chess platforms including Lichess Puzzles treat training as one undifferentiated puzzle stream with one rating. When a coach watches you play they don't say "you're a 1500." They say things like "your pattern recognition is solid but your calculation breaks down past three moves," or "you see the right ideas, you just don't prune candidates well — you burn five minutes calculating an irrelevant move." Those are different cognitive skills. They fail in different ways. They need different drills. A single rating can't tell them apart, which is why grinding more puzzles often plateaus your rating.

Climbchess separates training into four skills, each with its own dedicated mode and its own rating. Pattern recognition — seeing the right idea on the board — is trained in the open Puzzles stream over the SAE concept layer. Calculation — reading lines accurately N moves ahead — is trained in a Calculate mode that strips the position to the minimum number of pieces required to read the line, then scales piece count up as you improve, so you can train calculation in isolation without it being bottlenecked by your pattern recognition or visual memory simultaneously. Memory — holding a sequence in working memory — is trained in a Replay mode that shows position + sequence, hides it, asks you to reconstruct, and logs your span separately from your tactical rating. Move selection — choosing which two or three candidates are even worth deep calculation — is trained in a Move Selection mode where you pick candidates without calculation and the trainer tells you whether the strongest move was in your set.

After about fifty sessions you can see, by name, which of the four is bottlenecking your play. That's the diagnostic Lichess Puzzles can't give you because it's reporting one number. Coaches teach this; no other app cleanly splits the rating signal.

Feature comparison

Lichess Puzzles pricing at a glance

Free forever. Free tier: Yes — fully free, unlimited. Climbchess is free forever with no account required — that's the headline structural difference if cost is your decision driver.

What 30 days of Climbchess looks like

• Day 1: 12-rep session calibrates your concept profile. You see your starting per-concept ELO across 30+ patterns within 20 minutes.
• Day 7: Effort audit shows which 5 concepts you've over-trained vs the 3 critical gaps dragging your rating. Daily session auto-prioritises the gaps.
• Day 14: Critical-gap concepts up 80–150 ELO. Confidence on previously-uncomfortable positions visibly higher in your real games.
• Day 30: Per-concept ELO graph shows clear movement on the patterns you trained. You know — by name — the 3 patterns that were costing you points, and you've fixed them.

That's it. No subscription, no funnel. The trainer either moves your rating or it doesn't, and you'll see the data either way.

Best for vs where Climbchess wins

Lichess Puzzles is best for: Largest free puzzle set (4M+) with manual theme filtering.

Where Climbchess is different: Themes are categorical; no per-position explanation of which sub-pattern is the cue. Climbchess decomposes every puzzle into the specific neural-network features (4,505 of them) that fire on the position, so when you miss a puzzle you learn which pattern you didn't recognise — not just that you got it wrong.

The seven Climbchess training modes and what each one isolates

1. Today — daily 12-rep session tailored to your current concept profile and skill ratings. Mixes weak-concept drills with whichever of the four modes you've been underusing.
2. Curriculum — adaptive 30-day journey across principles, named tactics, and SAE patterns. Limited to the ~20 concepts that matter at your current band.
3. Puzzles — open SAE-tagged tactic stream. Trains pattern recognition.
4. Calculate — minimum-piece visualisation trainer. Trains calculation depth in isolation.
5. Replay — sequence memory trainer. Logs span separately from tactical rating.
6. Move Selection — candidate-pruning trainer. Trains the filter that runs before you start calculating.
7. Rating Test — bisects your true ELO in roughly ten minutes. No signup.

Plus an Endgame set of 80 curated studies with concept tags and solving guides, and real-game import: paste a PGN, the post-mortem extracts SAE concepts that fired on your mistakes and feeds them into the curriculum so the trainer learns what's costing you rating in actual games — not abstract puzzles.

Where Climbchess wins

• Four cognitive skills, four separate ratings. Pattern recognition, calculation, memory, and candidate-move selection each have their own dedicated mode and their own rating — so you can see which is bottlenecking your play. Lichess Puzzles reports a single rating.
• 4,505 SAE-derived concepts vs the ~30 manual themes most platforms use. Roughly 150× more granular pattern resolution, extracted automatically from a strong neural engine, not curated by hand.
• Per-concept ELO across every concept you've trained. The weakest one tells you exactly what's costing you points; the gap tells you how much.
• Effort audit — sorts concepts by (time spent) × (gap from your overall rating) so you can see what you've over-trained vs under-trained. Most players over-train what they're already good at; surfacing it explicitly is the diagnostic the app exists to provide.
• Calculate mode strips the position to the minimum number of pieces required to read the line, then scales piece count up as you improve. Lets you train calculation in isolation, not bottlenecked by pattern recognition or visual memory at the same time.
• Move Selection mode — pick two or three candidates without calculation; the trainer reveals whether the strongest move was in your set. Trains pruning, not the calculation that follows it.
• Replay mode — memory span trainer. Show position + sequence, hide it, reconstruct. Logs longest-sequence-repeated-correctly separately from your tactical rating.
• Real-game import + post-mortem. Paste a PGN, the trainer extracts SAE concepts that fired on your mistake positions and feeds them into the curriculum. Trains the patterns that are actually losing you rating in your games.
• Adaptive concept-level matchmaking. Puzzles are surfaced based on which concepts you're failing, not a flat difficulty slider.
• 9,950 SAE-tagged puzzles, 80 curated endgame studies, 562 concept-tagged opening drills — every position links to the concepts that fire on it. Opening drills are concept-tagged so you train pattern recognition on positions you'll actually reach, not move memorisation.
• 20-language UI including Hindi, Bengali, Tamil, Vietnamese, Indonesian, Arabic, Persian, Ukrainian — the long-tail Asian and RTL languages most platforms skip.
• Accessibility: colour-blind palettes (deuteranopia / protanopia / tritanopia / monochrome) with full UI flip, not just the chessboard. TTS pattern narration speaks concept name and outcome — useful while walking. Keyboard-first navigation throughout.
• Privacy posture: no ads, no third-party tracking, no analytics. Local-first storage. Optional anonymous cloud sync via UUID — no email, no account, no PII collected. SOC2-grade privacy posture.
• PWA: installable, offline-capable. Works on a flight, syncs when online.
• Open methodology with DOI — the SAE training pipeline, labelling, and matchmaking math are published. Read the paper. Every claim is reproducible. SAE pipeline and labelling code are open; trainer code currently closed.
• Free, no signup, no email. Take the rating test without an account.

Honest tradeoffs

• Newer than Lichess Puzzles — about 4 months old. The SAE pipeline is novel; battle-testing is ongoing.
• 9,950 curated positions vs unfiltered millions on big puzzle servers. We picked quality over volume — every position has a clean SAE-tagged solution. Power users grinding raw volume may prefer Lichess Puzzles.
• No live opponents or rated play. Climbchess is a trainer, not a chess server. Pair it with Lichess or Lichess Puzzles for actual games.
• Per-concept curriculum is noisy at the extremes — sub-1000 and above-2200 ELO — where puzzle density per concept drops.
• Some SAE concept labels are too narrow vs the underlying feature; we re-label periodically.
• A few concepts fire so rarely you can't get a stable per-concept rating on them.
• Calculate mode's piece-stripping heuristic occasionally removes a piece that mattered for the line.
• iOS Safari has rough mobile edges; Android and desktop are fine.

See also

• Climbchess vs Chess.com
• Climbchess vs Lichess
• Climbchess vs Chessable
• Climbchess vs Aimchess
• Climbchess vs ChessDojo
• Climbchess vs ChessTempo

For the underlying methodology see how it works, or jump into the free trainer.