Stockfish vs Reckless vs PlentyChess: Which Chess Engine Is Strongest Right Now?

Follow the major chess-engine rating lists through 2026 and one pattern keeps repeating: Stockfish, Reckless and PlentyChess trade the top three places, separated by a handful of Elo points — less than the lists' own statistical error bars. Stockfish is the long-reigning king. Reckless is a three-year-old challenger that has already reached the world championship final of engine chess. PlentyChess is the least famous of the three, yet it originated a technical idea that both of its rivals have since adopted. This article compares all three using sourced, date-stamped numbers as of November 2026 — and explains why the question "which one is strongest?" has no absolute answer.

Why all three stand at the top: the NNUE era

All three engines belong to the same school: classical alpha-beta search combined with an NNUE neural-network evaluation. NNUE (efficiently updatable neural networks) was born in Japanese shogi and brought to chess by Stockfish in 2020, and it quickly became the standard: instead of scoring a position with hundreds of hand-written rules, the engine learns to "see" positions from billions of self-played games, yet still runs fast on an ordinary CPU because each move only updates a small part of the network.

Once every strong engine uses NNUE, the race is about three subtler things: training data, network architecture, and search details — pruning, extensions, time management. The three engines here are three different answers to that same problem.

Stockfish — the long-reigning king

Stockfish is the oldest and largest project of the three. It descends from Tord Romstad's Glaurung engine (2004); the first Stockfish release appeared on 2 November 2008, developed by Romstad, Marco Costalba and Joona Kiiski. Today it is a global community project written in C++, released under the GPL-3.0 licence, with thousands of volunteers donating computing power to test every change through the Fishtest framework.

Its pace of progress has not slowed. In January 2026, Stockfish 18 added a new class of network features called "threat inputs" (see the PlentyChess section below) and gained up to 46 Elo over Stockfish 17 in official testing. On 5 September 2026, Stockfish 19 arrived, up to 44 Elo stronger again: it moved to the SFNNv16 network architecture, removed redundant threat features, added pawn-pair features and retired the secondary network used by earlier versions. Even so, development builds from late October 2026 have already moved on to the SFNNv19 architecture.

The record speaks for itself: according to Wikipedia (updated October 2026), Stockfish has won every main event of both TCEC and the Chess.com Computer Chess Championship (CCC) since 2020. At TCEC Season 29 — the unofficial world championship of engine chess, held in March–April 2026 — Stockfish won the 100-game Superfinal against Reckless by 59–41, collecting its 19th TCEC title and its 12th consecutive season crown. On the CCRL 40/15 list (30 October 2026 edition, 4 CPU), Stockfish 19 rates 3649 Elo, at the head of the established engines.

One little-noticed detail: SPCC's EAS list — which measures "aggressiveness" (sacrifices, fast wins, few bland draws) rather than strength — ranks Stockfish builds first and third. That is a consequence of overwhelming strength: Stockfish wins quickly and dares to sacrifice because it calculates further than anyone else. For players, Stockfish is also the default choice everywhere: free, available on every operating system, and the engine behind most popular analysis tools.

Reckless — the challenger written in Rust

Reckless is the most spectacular rise in recent engine chess. Its main author is Arseniy Surkov; the engine is written in Rust — a rarity in a C++-dominated field — and released under the AGPL-3.0 licence. Its first release, 0.1.0 in May 2023, scored only about 2005 Elo on the CCRL Blitz list, roughly strong-club-player level. Three years later, it is second in the world.

The turning point came with version 0.5.0 (February 2024): Reckless dropped its handcrafted evaluation and switched entirely to NNUE, with networks trained using the Bullet trainer on self-generated data. Its scoring has two personal touches: a correction-history mechanism, and a material-based "optimism" factor — the engine is slightly more confident in rich positions, a kind of character baked into the formula.

The project's own README lists these results for version 0.9.0 (released 1 March 2026): #2 on SPCC with 3833 Elo, #2 on CCRL Blitz with 3767, and #2 on CCRL 40/15 with 3612 (single-CPU list). On CCRL's 4-CPU 40/15 list (30 October 2026), Reckless 0.9.0 rates 3644 — just five points behind Stockfish 19.

But Reckless's biggest mark came at TCEC Season 29: second in the Premier Division, the top league, then into the Superfinal against Stockfish — breaking the Stockfish–Lc0 duopoly that had dominated that final for years. Reckless lost 41–59, but the match left a fair-play story recorded by Chessdom: after 16 games, the teams discovered a Rust compiler bug affecting Reckless, and jointly agreed to restart the entire Superfinal. On the EAS aggressiveness list, a Reckless development build ranks seventh and sits in the top five for sacrifice frequency — a personality worth consulting when you want a second opinion that differs from Stockfish.

PlentyChess — small team, and the gift of "threat inputs"

PlentyChess is essentially a one-author project: Patrick Leonhardt (online name Yoshie2000), written in C++, GPL-3.0, with its code repository opened in late 2023. Judging by team size alone, it should not be sitting at the same table as Stockfish — until you look at its technical contribution.

PlentyChess is where "threat inputs" were perfected and proven at the highest level: beyond piece placement, the network also receives information about which pieces are under threat and which squares are attacked — updated incrementally with every move and combined with king-position buckets. Its network was trained on more than 15 billion self-generated positions, covering both standard chess and Chess960, with self-distillation used in part of the training. The approach worked so well that Stockfish adopted it in Stockfish 18 (January 2026) and Reckless followed — a rare case where the smallest project became the teacher. The Chess Programming Wiki also credits Leonhardt with proposing a "voting on multiple moves" scheme for parallel search, a small but practical improvement in how many engines use multiple threads today.

For strength, the project's official ranking summary (updated November 2025) records: #3 on SPCC with 3789 Elo, #3 on CEGT with 3603, and #2 on CCRL with 3642. On CCRL's 4-CPU 40/15 list (30 October 2026), PlentyChess 7.0.0 rates 3642 and version 8.0.0 rates 3638 — every number within a few Elo of its two big rivals. At TCEC Season 29 it finished fourth in the Premier Division, behind Stockfish, Reckless and Lc0, and it is also a regular at the CCC. On the EAS list, PlentyChess 8.0.0 ranks sixth — neck and neck with Reckless — and is in the top five for fast wins.

Quick comparison table

• Team: Stockfish — global community, roots in Glaurung (2004), first release 2008 · Reckless — Arseniy Surkov and contributors, first release 2023 · PlentyChess — essentially Patrick Leonhardt alone, from late 2023
• Language / licence: Stockfish — C++ / GPL-3.0 · Reckless — Rust / AGPL-3.0 · PlentyChess — C++ / GPL-3.0
• Evaluation: Stockfish — NNUE, SFNNv16 architecture (version 19) · Reckless — NNUE + correction history + optimism factor · PlentyChess — NNUE with threat inputs, trained on 15B+ positions
• CCRL 40/15, 4 CPU (30 Oct 2026): Stockfish 19 — 3649 · Reckless 0.9.0 — 3644 · PlentyChess 7.0.0 — 3642
• SPCC (projects' official summaries): Stockfish — development builds top the list · Reckless — #2, 3833 · PlentyChess — #3, 3789
• TCEC Season 29 (2026): Stockfish — champion · Reckless — runner-up (lost the Superfinal 41–59) · PlentyChess — 4th in the Premier Division

One warning before reading too much into this table: each rating list is its own frame of reference — SPCC's 3833 (played with unbalanced UHO openings at fast time controls) cannot be compared directly with CCRL's 3644 (40 moves in 15 minutes on 4 CPUs). A number only means something together with its list, its date and its playing conditions.

Head-to-head: numbers that "lie"

The only long-time-control summit meeting between two of these three engines in 2026 was the TCEC Season 29 Superfinal: Stockfish beat Reckless 59–41 over 100 games. That is a clear margin but not a crushing one — and remember that TCEC deliberately uses unbalanced opening positions to force the engines to fight rather than draw quickly, so this score does not reflect what would happen from the standard starting position.

Change the arena and the story changes. In the Chess Engines Diary league (CEDR), season 10/2026, Stockfish again finished first, but a Reckless development build placed third and PlentyChess 8.0.0 only seventh in the same division. Even the CCRL list of 30 October 2026 has a brand-new name nominally above all three at 3650 — but on just 214 games, with error bars of ±28, far too small a sample to conclude anything. Read engine rating lists the way you read a weather forecast: right about the trend, always with probabilities attached.

Which engine should you analyse with?

For most players the answer is still Stockfish: strongest by most of the evidence, the most stable, the best supported, and present in nearly every analysis tool. If you install only one engine, install Stockfish.

Reckless and PlentyChess earn their place as a "second opinion". Top engines do not always agree: in the same position, one may choose a quiet positional move while another spots an interesting attacking idea. When two of these three names disagree, that is usually a sign the position is rich in ideas and worth digging into. Both are free, with ready-made builds for Windows, Linux and macOS (PlentyChess also builds for Android), plug into any UCI-compatible GUI, and since October 2026 Chessify has also added both to its cloud analysis platform. Chess960 players should keep an extra eye on PlentyChess, whose network was trained on 960 positions from the start.

And one honest truth: from a human player's perspective, a 5–20 Elo gap between these engines changes almost nothing about the quality of your analysis. The CPU you give the engine, and the time you let it think, matter far more than whether it ranks first or third.

Conclusion

As of November 2026, Stockfish remains the strongest chess engine by most of the evidence — titles, consistency across rating lists, and victory in the year's longest summit match. But the throne is no longer distant: Reckless, a three-year-old Rust project, has reached the TCEC Superfinal, and PlentyChess, nearly a one-person project, gave both of its rivals the threat-inputs idea that reshaped their network architectures. At today's summit, the gaps are measured in single Elo digits — and they shift from list to list and format to format.

A small note for SynerWorlds players: the engine running inside our own Chess game is Stockfish itself, tamed into 12 difficulty levels to fit human opponents. Understanding how strong that champion really is puts your games against it — games in which it is deliberately holding back — in a new light.

Sources: CCRL 40/15 list (30 October 2026 edition) and CCRL Blitz; SPCC rating list and EAS index (updated 14 October 2026); official README and ranking summaries of Reckless (March 2026) and PlentyChess (November 2025); Chessdom and the TCEC Wiki on Season 29; Chessify (20 October 2026); official Stockfish 18 and 19 release announcements.