{ }AICoder
2026 Live Verified

AI Coding Models Leaderboard

Aggregated official benchmark mirrors and real-world coding task scenarios. 100% verified, objective, and zero paid placement.

7Boards
87Models ranked
777Evaluation Tasks
August 2026Period
2026-08-24Last update
●We fetch raw rankings from 6 public benchmark providers directly — no re-weighting, no composite score. Click "Open official board" on each card to go to the source.

📌 7 official boards · mirrored 1:1

Each card previews a provider's Top 5. Click "View full" for the complete board, or jump to the official page. Data is 100% from each provider — we do no aggregation or math.

LMArena Agent ★
Human preference on real agentic coding sessions
Agent
Metric:Agent Score· 33 models
  1. #1Claude Opus 512.47
  2. #2Claude Fable 511.57
  3. #3Kimi K310.41
  4. #4GPT-5.6 Sol9.74
  5. #5Claude Opus 4.89.55
Official ↗
CursorBench
Cursor's first-party agentic coding benchmark (v3.2)
Agent
Metric:Pass rate%· 23 models
  1. #1Claude Fable 572.9%
  2. #2Grok 4.669.9%
  3. #3Gemini 3.7 Flash68.4%
  4. #4GPT-5.6 Sol67.2%
  5. #5Grok 4.566.7%
Official ↗
Artificial Analysis
Composite Intelligence Index over 20+ benchmarks
Overall
Metric:AA Index· 80 models
  1. #1Claude Opus 563.1
  2. #2Claude Fable 562.1
  3. #3GPT-5.6 Sol60.9
  4. #4Grok 4.660.9
  5. #5Kimi K359.7
Official ↗
LiveBench
Contamination-resistant benchmark, refreshed monthly
Reasoning
Metric:Avg score%· 32 models
  1. #1Claude Fable 583.4%
  2. #2GPT-5.6 Sol81.6%
  3. #3GPT-5.580.8%
  4. #4Claude Opus 580.5%
  5. #5Gemini 3.7 Flash79.9%
Official ↗
Aider Polyglot
133 real coding tasks across 6 languages
Coding
Metric:Pass rate%· 35 models
  1. #1GPT-5.6 Sol88.0%
  2. #2o3-pro84.9%
  3. #3Gemini 2.5 Pro83.1%
  4. #4o381.3%
  5. #5Grok 4.679.6%
Official ↗
LiveCodeBench
Continuously updated competitive-programming set
Reasoning
Metric:Pass@1%· 15 models
  1. #1DeepSeek-V3100.0%
  2. #2GPT-4o100.0%
  3. #3GPT-4o mini100.0%
  4. #4OpenAI o3-mini (High Reasoning Coding)100.0%
  5. #5o4-mini100.0%
Official ↗
EvalPlus
HumanEval+ strict test suite
Coding
Metric:HumanEval+%· 21 models
  1. #1o189.0%
  2. #2o1-mini89.0%
  3. #3Qwen3.6-Plus87.2%
  4. #4GPT-4o87.2%
  5. #5DeepSeek-V386.6%
Official ↗

LMArena AgentAgent

Metric: Agent Score·33 models·fetched 2026-08-24

Open official board ↗
#ModelAgent ScoreValue ScoreCross-board ranks
#1
Claude Opus 5Anthropic
12.47
5.6
C—
A #1
L #4
A—L—E—
#2
Claude Fable 5Anthropic
11.57
7.9
C #1
A #2
L #1
A—L—E—
#3
Kimi K3Moonshot AI
10.41
24.7
C #12
A #5
L #6
A—L—E—
#4
GPT-5.6 SolOpenAI
9.74
11.6
C #4
A #3
L #2
A #1
L—
E #21
#5
Claude Opus 4.8Anthropic
9.55
14.7
C #9
A #8
L #14
A #8
L #10
E #11
#6
GPT-5.5OpenAI
8.51
17.7
C #17
A #10
L #3
A #19
L—E—
#7
Claude Sonnet 5Anthropic
6.62
35
C #13
A #13
L #16
A—L—
E #8
#8
DeepSeek-V4-ProDeepSeek
6.26
139.5
C #10
A #14
L #11
A—L—E—
#9
Qwen3.8-MaxQwen
6.2
47.5
C #8
A #7
L #7
A—L—E—
#10
Grok 4.5xAI
6.17
46.9
C #5
A #12
L #15
A—L—E—
#11
GLM-5.2Zhipu AI
5.82
61
C #19
A #16
L #24
A—L—E—
#12
GPT-5.4OpenAI
4.98
32.9
C—
A #15
L #9
A—L—E—
#13
GPT-5.6 LunaOpenAI
4.04
291.7
C #14
A #17
L #22
A—L—E—
#14
DeepSeek-V4-FlashDeepSeek
3.99
395.2
C #15
A #20
L #19
A—L—E—
#15
Gemini 3.7 FlashGoogle
3.32
92.7
C #3
A #11
L #5
A—L—E—
#16
GPT-5.6 TerraOpenAI
3.19
30.5
C #7
A #9
L #10
A—L—E—
#17
Claude Sonnet 4.6Anthropic
2.88
21.2
C #21
A #22
L #23
A #13
L #9
E #15
#18
Kimi K2.7 CodeMoonshot AI
0.43
53.4
C #20
A #27
L #27
A—L—E—
#19
GLM-5.1Zhipu AI
0
52.9
C—
A #31
L—A—L—E—
#20
Gemini 3.5 FlashGoogle
-0.68
35.5
C #22
A #18
L #18
A—L—E—
#21
Qwen3.7-MaxQwen
-0.75
70.4
C—
A #24
L #21
A #15
L #13
E—
#22
Gemini 3.1 ProGoogle
-0.88
58.2
C—
A #23
L #12
A—L—E—
#23
Kimi K2.6Moonshot AI
-1.1
68.9
C #23
A #26
L #25
A #16
L—E—
#24
Gemini 3.6 FlashGoogle
-3.04
83.3
C—
A #21
L #20
A—L—E—
#25
MiniMax-M3MiniMax
-3.15
218.4
C—
A #25
L #29
A—L—E—
#26
Mistral Small 4Mistral AI
-6.82
633
C—
A #29
L—A—L—E—
#27
Mistral Medium 3.5Mistral AI
-7.6
96.8
C—
A #44
L—A—L—E—
#28
Grok 4.3xAI
-9.02
60.2
C—
A #36
L #32
A—L—E—
#29
Grok Build 0.1xAI
-9.48
79.2
C—
A #33
L #28
A—L—E—
#30
Gemini 2.5 ProGoogle
-10.82
37.3
C—
A #28
L—
A #3
L #11
E #10
#31
Gemini 3.5 Flash-LiteGoogle
-10.82
106.7
C—
A #38
L #31
A—L—E—
#32
MiniMax-M2.7MiniMax
-11.86
147.8
C—
A #35
L—A—L—E—
#33
Nemotron-UltraNvidia
-15.36
3.3
C—
A #50
L—A—L—E—
Rank color:#1#2#3#4-5#6-10#11+

The "Cross-board ranks" column shows this model's rank on the other boards at a glance (grey = not tested there). Boards use different methodologies, so raw values are not directly comparable.

LMArena Agent board: real agentic coding sessions scored by human blind votes (signal-based agent score). Sourced from arena.ai/leaderboard/agent and mirrored 1:1 — we recompute nothing.

📊 7 benchmark providers
LMArena Agent ↗ Agent

LMArena Agent board: real agentic coding sessions scored by human blind votes (signal-based agent score). Sourced from arena.ai/leaderboard/agent and mirrored 1:1 — we recompute nothing.

Agent Score33 models
CursorBench ↗ Agent

CursorBench is Cursor's first-party agentic coding benchmark (v3.2). Tasks come from real Cursor sessions with ambiguous, multi-file requirements. Metric: pass rate (%). Mirrored from Cursor's official cursor.com/cursorbench snapshot.

Pass rate%23 models
Artificial Analysis ↗ Overall

Artificial Analysis aggregates 20+ benchmarks (MMLU-Pro, GPQA, HLE, SciCode, IFBench, Terminal-bench, etc.) into a single Intelligence Index — the most comprehensive commercial evaluator for "overall capability".

AA Index80 models
LiveBench ↗ Reasoning

LiveBench refreshes its questions monthly to prevent contamination. Covers reasoning, math, coding, language, instruction following, and data analysis.

Avg score%32 models
Aider Polyglot ↗ Coding

133 real coding tasks (bug-fix / feature) across 6 languages. Metric: pass_rate_2.

Pass rate%35 models
LiveCodeBench ↗ Reasoning

Continuously updated competitive-programming problems. Metric: Pass@1.

Pass@1%15 models
EvalPlus ↗ Coding

HumanEval+ strict test suite, 164 tasks.

HumanEval+%21 models

All rankings and numbers come directly from official sources; we only do name matching and aggregate presentation, no weighting or re-ranking. Boards use different methodologies, so do not compare raw values across boards.