AI coding benchmark leaderboard (November 2026)
As of 13 Nov 2026, BenchLeader tracks 44 AI coding benchmarks from 17 publishers, 13 of which feed the BenchLeader Index. Across coding benchmarks, Claude Fable 5.2 (Anthropic) leads with a coding score of 80.8, ahead of Claude Opus 5.5 (79.8) and Gemini 4 Carbon (74.1). On the most-weighted live board, Arena WebDev is led by Claude Fable 5.2 (xhigh) at 1838. Of the 44 boards, 30 are active, 9 near-saturated, 3 saturated and 2 retired (retired: no new result in a year).
Last updated · rebuilt every six hours · independent publishers only, no launch-post numbers
- benchmarks
- 44
- active
- 30
- saturated or near
- 12
- publishers
- 17
Other categories: AI agent benchmarks · Reasoning and math benchmarks · Multimodal AI benchmarks · all LLM benchmarks
Which models lead coding benchmarks
Ranked on the coding category of the BenchLeader Index: every board below that feeds the index, standardised (50 is the average model) and averaged. One row per model, at its best reasoning effort; models need results on at least 2 of these boards.
| # | Model | Provider | Coding score | Index | Blended $/M |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 (xhigh) | Anthropic | 80.8 | 69.9 | $16.00 |
| 2 | Claude Opus 5.5 (max) | Anthropic | 79.8 | 68.4 | $8.00 |
| 3 | Gemini 4 Carbon (high) | 74.1 | 69.5 | $4.00 | |
| 4 | Claude Fable 5 (max) | Anthropic | 73.6 | 65.0 | $20.00 |
| 5 | GPT-6 Astra (max) | OpenAI | 73.2 | 67.1 | $20.00 |
| 6 | Claude Fable 5.1 | Anthropic | 72.9 | 66.4 | $20.00 |
| 7 | GPT-5.6 Sol | OpenAI | 71.6 | 54.3 | $8.00 |
| 8 | Claude Sonnet 5.5 (max) | Anthropic | 71.5 | 64.7 | $4.00 |
| 9 | GPT-6.1 Sol (max) | OpenAI | 71.1 | 66.6 | $4.00 |
| 10 | Muse Wonder Full (max) | Meta | 69.9 | 61.7 | – |
Every coding benchmark, what it measures and who leads (44)
Coding benchmarks range from contest problems published after a model's training cut-off (LiveCodeBench) to real GitHub issues fixed end to end and checked by the repository's own tests (SWE-bench Verified, 500 issues; SWE-Bench Pro in private repositories) and edits made through a real coding agent (Aider Polyglot, 225 exercises).
Use them for: choosing a model for code generation, review and autonomous coding agents.
Caveat: HumanEval is saturated, and on SWE-bench the harness and scaffold move scores as much as the model does: compare like with like (BenchLeader shows the bash-only harness separately).
| Benchmark | What it measures | Leader today | Status | Models | Newest result |
|---|---|---|---|---|---|
| Arena WebDev | Pairwise votes on generated web apps. | Claude Fable 5.2 (xhigh) 1838 | ActiveJudge-based | 150 | 11 Nov 2026 |
| LiveCodeBench | Competitive programming problems released after model cutoffs. Run by Vals AI. | Claude Fable 5.1 90.5% | Near-saturated | 141 | 1 Sep 2026 |
| FrontierCode 2.0 (AA) | Cognition's FrontierCode 2.0: 132 real open-source repository tasks graded on whether a maintainer would merge the… | Claude Fable 5.2 (high) 49.1% | Active | 133 | 12 Nov 2026 |
| SWE-bench Verified (bash only) | Official SWE-bench Verified results from the minimal bash-only mini-SWE-agent harness. | Claude Opus 4.5 (high) 76.8% | ActiveHarness-dependent | 46 | 26 Feb 2026 |
| SWE-Bench Pro | Long-horizon software engineering tasks in public repositories. Scale AI. | Muse Spark 1.1 61.5% | ActiveHarness-dependent | 25 | 9 Jul 2026 |
| Arena Coding | Text arena rating on coding prompts. | Gemini 4 Carbon (high) 1568 | ActiveJudge-based | 379 | 11 Nov 2026 |
| SciCode | Research-grade scientific coding problems. | Claude Fable 5.2 (max) 67.7% | Active | 236 | 10 Nov 2026 |
| WeirdML | Unusual machine-learning tasks that require writing working training code. | GPT-6 Astra (pro mode) 93.6% | Near-saturated | 159 | 3 Sep 2026 |
| FrontierCode | Hard real-world coding tasks run in agent harnesses (Cognition). | Claude Opus 5.5 (medium) 54.6% | Active | 46 | 13 Oct 2026 |
| GSO-Bench | Software optimisation tasks measured against expert speed-ups. | Claude Fable 5.1 88.2% | Active | 35 | 3 Sep 2026 |
| SWE-bench Verified (Epoch) | Epoch AI's own SWE-bench Verified runs with a constant scaffold. | Claude Opus 4.7 (max) 83.5% | ActiveHarness-dependent | 33 | 16 Jun 2026 |
| LiveCodeBench (AA)not in index | Competitive programming problems harvested continuously from LeetCode, AtCoder and Codeforces after a model's training cut-off, so… | Gemini 3 Pro (high) 91.8% | Near-saturated | 307 | 4 Jan 2026 |
| Terminal-Bench 4.0 (AA)not in index | Real command-line tasks in a sandboxed terminal, version 4.0. Run by Artificial Analysis; the Terminal-Bench team's… | Kimi K3.1 Pro 67.2% | ActiveHarness-dependent | 257 | 12 Nov 2026 |
| SciCode (AA)not in index | Artificial Analysis' own SciCode run. Removed from Intelligence Index v5; still published. Shown for reference. | Claude Fable 5.2 (max) 67.7% | Active | 251 | 9 Nov 2026 |
| Terminal-Bench 2.1 (AA)not in index | The previous Terminal-Bench release, kept because far more models have been run on it than on… | Claude Fable 5.1 (max) 91.4% | Near-saturatedHarness-dependent | 240 | 5 Nov 2026 |
| Design Arena Websitenot in index | Two models build a website from the same prompt, the visitor votes for the better one… | Muse Wonder (max) 1378 | ActiveJudge-based | 193 | 13 Nov 2026 |
| GameBench 2 by reasoning settingnot in index | GameBench 2 scored separately for each reasoning setting a model was run at, filed under the… | Claude Opus 5.5 (xhigh) 94.9 | Active | 189 | 10 Nov 2026 |
| ALE-Benchnot in index | Long-horizon algorithm engineering on AtCoder heuristic contests. | GPT-6 Astra (max) 2951.3 | Active | 141 | 9 Nov 2026 |
| CursorBenchnot in index | Cursor's internal coding-agent evaluation, with reasoning level and cost. | Gemini 4 Carbon (high) 60.4% | ActiveHarness-dependent | 90 | 10 Nov 2026 |
| FrontierCode 1.1 (AA)not in index | Artificial Analysis' run of Cognition's FrontierCode 1.1. Kept beside 2.0 because far more models have been… | Claude Fable 5.2 (medium) 64.0% | Active | 83 | 9 Nov 2026 |
| Code Migrationnot in index | Rewriting whole programs in another language so they still pass the tests. Run by Vals AI. | Claude Sonnet 5.5 69.8% | Active | 78 | 29 Oct 2026 |
| DeepSWE v1.1not in index | Software-engineering tasks in an agent harness with reasoning effort recorded. | GPT-6 Astra (xhigh) 74.1% | ActiveHarness-dependent | 69 | 3 Sep 2026 |
| LiveBench 1.0 Codingnot in index | Coding category of LiveBench 1.0, frozen at its final release (25 June 2026). LiveBench 2.0's Coding… | Claude Sonnet 5.5 (max) 91.4% | Near-saturated | 68 | 25 Jun 2026 |
| SWE-bench Verified (any scaffold)not in index | Best official SWE-bench Verified result for the model with any agent scaffold. Shown for reference. | Claude Opus 4.5 79.2% | ActiveHarness-dependent | 66 | 19 Feb 2026 |
| ProgramBenchnot in index | Rebuilding programs from compiled binaries. Run by Vals AI. | Claude Fable 5.2 19.0% | Active | 65 | 9 Nov 2026 |
| GameBench 2 Overallnot in index | DuelLab's overall GameBench 2 score: models write game-playing programs, the programs play each other, and the… | GPT-6 Astra 88.6 | Active | 52 | 10 Nov 2026 |
| MirrorCodenot in index | Reproducing a program's behaviour from its outputs. Run by Epoch AI. | Claude Sonnet 5.5 (max) 97.8% | Near-saturated | 45 | 26 Oct 2026 |
| Startrise Frontendnot in index | Twelve single-shot frontend briefs, built in a real browser and scored on craft, engineering, brief adherence… | Kimi K3.1 Pro 82.5% | Near-saturated | 37 | 28 Oct 2026 |
| LiveBench 2.0 Codingnot in index | LiveBench 2.0 coding suite: real merged-PR bug fixes on a frozen monorepo, graded deterministically. | Gemini 4 Carbon 91.0% | Near-saturated | 34 | 6 Nov 2026 |
| Code Migration v1.1not in index | Code Migration v1.1, the version Vals AI now runs new models on. Run by Vals AI. | Claude Sonnet 5.5 86.9% | Active | 30 | 9 Nov 2026 |
| Surface Evolver Benchnot in index | Scientific computing with the Surface Evolver package. | Claude Fable 5 (high) 95.0% | Near-saturated | 29 | 6 Oct 2026 |
| Vibe Code Bench 1-100not in index | Extending a working web app, request by request, one hundred times. Run by Vals AI. | Muse Wonder Full (max) 34.0% | Active | 26 | 29 Oct 2026 |
| SWE-Bench Pro (private)not in index | The held-out commercial-repository set of SWE-Bench Pro. Scale AI. | Muse Spark 1.1 61.5% | ActiveHarness-dependent | 25 | 9 Jul 2026 |
| FrontierSWEnot in index | Frontier-difficulty software-engineering tasks across implementation, performance and research. | Claude Fable 5.2 (max) 69.9% | ActiveHarness-dependent | 22 | 10 Nov 2026 |
| SWE Atlas: Codebase QnAnot in index | Answering questions about a large codebase. Scale AI. | Claude Opus 5 (xhigh) 63.2% | ActiveHarness-dependent | 22 | 9 Sep 2026 |
| SWE Atlas: Test Writingnot in index | Writing tests for existing code. Scale AI. | Claude Fable 5.1 (xhigh) 67.0% | ActiveHarness-dependent | 22 | 9 Sep 2026 |
| SWE-bench Multimodalnot in index | GitHub issues from JavaScript repositories whose bugs are visual, so fixing them means reading images. Run… | Claude Opus 5.5 70.2% | ActiveHarness-dependent | 22 | 9 Nov 2026 |
| AlgoTunenot in index | Speeding up numeric code beyond reference implementations. | GPT-5.2 (medium) 2.0 | Active | 18 | 5 Mar 2026 |
| SWE Atlas: Refactoringnot in index | Refactoring tasks in real repositories. Scale AI. | GPT-6 Astra (xhigh) 59.0% | ActiveHarness-dependent | 17 | 9 Sep 2026 |
| IOI | International Olympiad in Informatics problems. Run by Vals AI. | GPT-6 Astra (max) 100.0% | Saturated | 46 | 29 Oct 2026 |
| Vibe Code Bench v1.1not in index | Building web apps from scratch from a brief. Run by Vals AI. | Claude Fable 5.2 93.8% | SaturatedJudge-based | 126 | 11 Nov 2026 |
| SWE-bench (Vals)not in index | Vals AI's own SWE-bench run. Shown for reference. | Claude Opus 5 97.0% | SaturatedHarness-dependent | 87 | 1 Sep 2026 |
| Aider Polyglot | Percent of 225 Exercism problems solved on the second attempt. | GPT-5 88.0% | RetiredHarness-dependent | 49 | 3 Oct 2025 |
| CadEvalnot in index | Generating CAD models from specifications. | o3 (medium) 74.0% | Retired | 15 | 16 Apr 2025 |
AI coding benchmarks: questions people ask
- What are the best AI coding benchmarks?
- The most informative AI coding benchmarks today are Arena WebDev (led by Claude Fable 5.2 (xhigh) at 1838), LiveCodeBench (led by Claude Fable 5.1 at 90.5%) and FrontierCode 2.0 (led by Claude Fable 5.2 (high) at 49.1%): each feeds the BenchLeader Index, covers many models and has not been saturated. Use them for: choosing a model for code generation, review and autonomous coding agents.
- Which AI model is best on AI coding benchmarks?
- As of 13 Nov 2026: Across coding benchmarks, Claude Fable 5.2 (Anthropic) leads with a coding score of 80.8, ahead of Claude Opus 5.5 (79.8) and Gemini 4 Carbon (74.1). A category score standardises every board (50 is the average model, 15 points one standard deviation) and averages them, so one board cannot decide it.
- Are AI coding benchmarks comparable across boards?
- Partly. Contest-style boards (LiveCodeBench, SciCode) grade the code itself and compare well across models; repository boards (SWE-bench Verified, SWE-Bench Pro) also depend on the agent harness that runs the model, so compare results from the same harness. BenchLeader keeps each harness as its own board.
- Which AI coding benchmarks are saturated?
- As of 13 Nov 2026, LiveCodeBench (near-saturated), WeirdML (near-saturated), LiveCodeBench (near-saturated), Terminal-Bench 2.1 (near-saturated), LiveBench 1.0 Coding (near-saturated), MirrorCode (near-saturated), Startrise Frontend (near-saturated) and LiveBench 2.0 Coding (near-saturated) and 4 more. A board is saturated when its fifth-best model scores 90% or more and near-saturated when the leader reaches 90% or the fifth-best 80%; both are decided from the data and change as scores do.
- How often is this leaderboard updated?
- BenchLeader re-reads every publisher's results every six hours (runs start at 04:03, 10:03, 16:03 and 22:03 UTC), and this page is rebuilt with each update. Each board's newest result date is in the table; publishers update on their own schedules.
Related
Cite as: BenchLeader, “AI coding benchmark leaderboard (November 2026)”, data as of 13 Nov 2026, https://www.benchleader.com/benchmarks/coding.