Which model leads on Artificial Analysis · Coding Index?

Claude Fable 5 from Anthropic leads Artificial Analysis · Coding Index with a score of 76.5. The median score across 89 tested models is 35.5.

Is Artificial Analysis · Coding Index saturated?

Yes · the top model on Artificial Analysis · Coding Index has reached 76.5 out of 60, within 5% of the theoretical ceiling. This benchmark is approaching saturation and may be replaced by a harder successor.

Does Artificial Analysis · Coding Index predict performance on other benchmarks?

Yes · Artificial Analysis · Coding Index scores correlate 0.96 with HELM · WildBench across 5 shared models. Models that do well on Artificial Analysis · Coding Index tend to do well on HELM · WildBench.

How often is Artificial Analysis · Coding Index data refreshed?

BenchGecko pulls updates daily. New model scores on Artificial Analysis · Coding Index appear as soon as they are published by Epoch AI or the model provider.

Benchmark · KnowledgeSettled

Artificial Analysis · Coding Index

Name: Artificial Analysis · Coding Index Benchmark
Creator: BenchGecko
License: https://creativecommons.org/licenses/by/4.0/

Artificial Analysis Coding Index · a composite score that aggregates performance across multiple coding benchmarks into a single index. Tracks code generation quality, debugging ability, multi-language competence, and real-world software engineering tasks. Used by Artificial Analysis to rank model coding capability in a normalized, comparable format. Useful for developers choosing between models for coding-heavy workloads.

Updated 2026-06-16

The composite index reveals that coding capability has become the most contested dimension among frontier models. The top-5 spread is consistently tighter than on any single coding benchmark alone.

Scoring: Weighted composite of multiple coding benchmark scores. Higher is better. Scale varies by evaluation cycle.

Models tested

Top score

76.5

Claude Fable 5

Median

35.5

min 0.8

Top-5 spread

σ 2.8

Competitive

The Frontier

Best score over time · one chart, every benchmark

Chart type

Frontier on Artificial Analysis · Coding Index rose from 11.2 to 76.5 in 17 months · +65.3 points · latest leader Claude Fable 5 from Anthropic.

Pink dots = frontier records · 11 totalClick to open model page

Full rankings

89 models tested · sorted by score

#	Model	Score	Price
1	Claude Fable 5· Anthropic	76.5	$10.00
2	Claude Opus 4.8 (Fast)· Anthropic	74.3	$10.00
3	Claude Opus 4.7 (Fast)· Anthropic	73.6	$30.00
4	Gemini 3.5 Flash· Google DeepMind	70.1	$1.50
5	Gemini 3.1 Pro Preview· Google DeepMind	68.8	$2.00
6	GLM 5.2· z-ai	68.8	$1.00
7	Qwen3.7 Max· Alibaba Qwen	66.0	$1.25
8	Claude Sonnet 4.6· Anthropic	63.0	$3.00
9	Kimi K2.7 Code· moonshotai	60.8	$0.61
10	MiMo-V2.5-Pro· xiaomi	60.2	$0.43
11	DeepSeek V4 Pro· DeepSeek	59.4	$0.43
12	U Muse Spark· Unknown	58.6	—
13	MiniMax M3· minimax	58.6	$0.30
14	GPT-5.4· OpenAI	57.3	$2.50
15	DeepSeek V4 Flash· DeepSeek	56.2	$0.08
16	GPT-5.4 Mini· OpenAI	56.1	$0.75
17	GPT-5.4 Nano· OpenAI	56.1	$0.20
18	Kimi K2.6· moonshotai	56.0	$0.66
19	Qwen3.7 Plus· Alibaba Qwen	55.9	$0.32
20	GLM 5.1· z-ai	55.8	$0.97
21	Qwen3.6 Plus· Alibaba Qwen	54.5	$0.33
22	Qwen3.6 27B· Alibaba Qwen	53.7	$0.28
23	GPT-5.3-Codex· OpenAI	53.1	$1.75
24	MiniMax M2.7· minimax	52.6	$0.24
25	Nemotron 3 Ultra· NVIDIA	49.3	$0.50
26	Qwen3.5 397B A17B· Alibaba Qwen	48.2	$0.39
27	Claude Opus 4.6 (Fast)· Anthropic	48.1	$30.00
28	Mistral Medium 3.5· Mistral AI	46.9	$1.50
29	Qwen3.5-122B-A10B· Alibaba Qwen	45.7	$0.26
30	GLM 5 Turbo· z-ai	44.2	$1.20
31	Gemma 4 31B (free)· Google DeepMind	43.4	$0.00
32	I Ring-2.6-1T· inclusionai	42.8	$0.07
33	Gemini 3 Flash Preview· Google DeepMind	42.6	$0.50
34	Grok 4.3· xAI	42.3	$1.25
35	Qwen3.6 35B A3B· Alibaba Qwen	41.9	$0.14
36	MiMo-V2-Pro· xiaomi	41.4	$1.00
37	Kimi K2.5· moonshotai	39.5	$0.38
38	Gemini 3 Pro· Google DeepMind	39.4	—
39	Gemma 4 26B A4B (free)· Google DeepMind	39.3	$0.00
40	o3· OpenAI	38.4	$2.00
41	DeepSeek V3.2 Speciale· DeepSeek	37.9	$0.40
42	Step 3.7 Flash· stepfun	37.3	$0.20
43	DeepSeek V3.2· DeepSeek	36.7	$0.21
44	GLM 5V Turbo· z-ai	36.2	$1.20
45	MiMo-V2-Omni· xiaomi	35.5	$0.40
46	Qwen3.5-27B· Alibaba Qwen	34.9	$0.20
47	Gemini 3.1 Flash Lite· Google DeepMind	34.7	$0.25
48	MiMo-V2-Flash· xiaomi	33.5	$0.09
49	Gemini 2.5 Pro· Google DeepMind	31.9	$1.25
50	Step 3.5 Flash· stepfun	31.6	$0.10
51	Grok 4.1 Fast· xAI	30.9	$0.20
52	Mercury 2· inception	30.6	$0.25
53	Qwen3 Max Thinking· Alibaba Qwen	30.5	$0.78
54	gpt-oss-120b (free)· OpenAI	30.4	$0.00
55	Qwen3.5-35B-A3B· Alibaba Qwen	30.3	$0.14
56	Gemini 3.1 Flash Lite Preview· Google DeepMind	30.1	$0.25
57	Qwen3.5-9B· Alibaba Qwen	28.7	$0.10
58	Trinity Large Thinking· arcee-ai	27.2	$0.25
59	Qwen3 Coder 480B A35B (free)· Alibaba Qwen	24.6	$0.00
60	Mistral Small 4· Mistral AI	24.3	$0.15
61	R1 0528· DeepSeek	24.0	$0.50
62	Grok Code Fast 1· xAI	23.7	$0.20
63	Qwen3 Coder Next· Alibaba Qwen	22.9	$0.11
64	Qwen3.5 4B· Alibaba	22.6	—
65	gpt-oss-20b (free)· OpenAI	20.7	$0.00
66	Qwen3.5 2B· Alibaba	19.6	—
67	Qwen3 Next 80B A3B Instruct (free)· Alibaba Qwen	19.5	$0.00
68	INTELLECT-3· prime-intellect	19.1	$0.20
69	Mistral Medium 3.1· Mistral AI	18.3	$0.40
70	Gemini 2.5 Flash Lite· Google DeepMind	18.1	$0.10
71	Llama 4 Maverick· Meta	16.3	$0.15
72	Solar Pro 3· upstage	16.2	$0.15
73	Qwen3 Next 80B A3B Instruct· Alibaba Qwen	15.3	$0.09
74	Qwen3.5 0.8B· Alibaba	15.0	—
75	ERNIE 4.5 300B A47B · baidu	14.5	$0.28
76	Llama 3.1 Nemotron Ultra 253B v1· NVIDIA	13.1	$0.60
77	R1 Distill Llama 70B· DeepSeek	11.4	$0.80
78	Phi 4· Microsoft	11.2	$0.07
79	Granite 4.1 8B· ibm-granite	10.3	$0.05
80	Command A· Cohere	9.9	$2.50
81	Reka Flash 3· rekaai	8.9	$0.10
82	N Nanbeige4.1 3B· Nanbeige	8.9	—
83	NVIDIA Nemotron Nano 9B V2· NVIDIA	8.3	—
84	Llama 4 Scout· Meta	8.2	$0.10
85	Granite 4.0 Micro· ibm-granite	5.0	$0.02
86	LFM2-24B-A2B· liquid	3.6	$0.03
87	Phi 4 Mini Instruct· Microsoft	3.6	—
88	LFM2.5-1.2B-Thinking (free)· liquid	1.4	$0.00
89	LFM2.5-1.2B-Instruct (free)· liquid	0.8	$0.00

Details

Category: Knowledge
Creator: Artificial Analysis
Max score: 60
Modality: Code
Scoring: Weighted composite of multiple coding benchmark scores. Higher is better. Scale varies by evaluation cycle.
Models: 89
Updated: 2026-06-16

Tests

Code generationDebuggingMulti-language codingSoftware engineering

Does not test

VisionLong contextSafetyScientific reasoning

Links

Artificial Analysis

Gecko's Take

“Composite indices are useful for quick comparison, but always drill into the components. A model with a high AA Coding Index might excel at HumanEval but struggle on SWE-bench. Trust the components for production decisions.”

Related benchmarks