Benchmark suiteLatest source data: Sep 1, 2026Checked: September 2, 2026

Coding model rankings.

A focused coding view using Code Arena and SWE-bench today, with clear source slots for Terminal-Bench and Aider-style coding benchmarks as those feeds are wired.

What These Mean

Ranked models

101

Code Arena or SWE-bench signal

SWE-bench matches

Matched to model rows

Top coding index

Qwen3.8 Max 0902

100 coding index

Best SWE-bench

Pending

Coding formula

Code generation plus real issue resolution.

The coding index blends Code Arena and SWE-bench where available, with output speed as a small tie-breaker. Terminal-Bench and Aider remain visible as planned feed slots until we have a reliable ingestion path.

Coding performance

Higher coding index is better. SWE-bench cells show verified model matches only.

Top 30 of 101

Rank	Model	Index	Code Arena	SWE-bench	Speed	Run mode	Sources
#1	Qwen3.8 Max 0902 qwen3.8-max-0902 AlibabaProprietary	100coding index	1,691Code Arena rank #1	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-bench
#2	Claude Opus 5 Max claude-opus-5-max AnthropicProprietary	99coding index	1,688Code Arena rank #2	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-bench
#3	Kimi K3 Max kimi-k3-max Moonshot AIProprietary	95coding index	1,674Code Arena rank #3	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-bench
#4	Hy4 Preview hy4-preview TencentOpen weights	83coding index	1,629Code Arena rank #6	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-bench
#5	Qwen3.8 Max qwen3.8-max AlibabaProprietary	78coding index	1,669Code Arena rank #4	Not listedNo confident match	40 tok/soutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#6	GLM 5.3 Max glm-5.3-max Z.aiOpen weights	77coding index	1,608Code Arena rank #11	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-bench
#7	Claude Opus 5 High claude-opus-5-high AnthropicProprietary	77coding index	1,661Code Arena rank #5	Not listedNo confident match	47 tok/soutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#8	Gemini 3.7 Flash High gemini-3.7-flash-high GoogleProprietary	74coding index	1,587Code Arena rank #14	Not listedNo confident match	286 tok/soutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#9	GLM 5.2 Max glm-5.2-max Z.aiOpen weights	71coding index	1,585Code Arena rank #15	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-bench
#10	DeepSeek V4 Pro High 20260813 deepseek-v4-pro-high-20260813 DeepSeekOpen weights	71coding index	1,584Code Arena rank #16	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-bench
#11	DeepSeek V4 Flash High deepseek-v4-flash-high DeepSeekOpen weights	70coding index	1,581Code Arena rank #17	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-bench
#12	Claude Fable 5 claude-fable-5 AnthropicProprietary	70coding index	1,628Code Arena rank #8	Not listedNo confident match	59 tok/soutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#13	Qwen3.8 Flash Next qwen3.8-flash-next AlibabaProprietary	70coding index	1,620Code Arena rank #9	Not listedNo confident match	84 tok/soutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#14	Grok 4.6 High grok-4.6-high xAIProprietary	70coding index	1,629Code Arena rank #7	Not listedNo confident match	50 tok/soutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#15	GPT-5.6 Sol xHigh gpt-5.6-sol-xhigh OpenAIProprietary	68coding index	1,616Code Arena rank #10	Not listedNo confident match	76 tok/soutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#16	Claude Opus 4.8 High claude-opus-4-8-high AnthropicProprietary	65coding index	1,563Code Arena rank #18	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-bench
#17	Claude Opus 4.7 claude-opus-4-7 AnthropicProprietary	64coding index	1,557Code Arena rank #19	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#18	GLM 5.3 Flash glm-5.3-flash Z.aiOpen weights	64coding index	1,604Code Arena rank #12	Not listedNo confident match	42 tok/soutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#19	Claude Opus 4.7 High claude-opus-4-7-high AnthropicProprietary	63coding index	1,556Code Arena rank #20	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-bench
#20	Qwen3.8 27b qwen3.8-27b AlibabaOpen weights	63coding index	1,599Code Arena rank #13	Not listedNo confident match	49 tok/soutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#21	Claude Opus 4.6 High claude-opus-4-6-high AnthropicProprietary	61coding index	1,546Code Arena rank #22	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-bench
#22	Muse Spark 1.1 muse-spark-1.1 MetaProprietary	59coding index	1,540Code Arena rank #24	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#23	Claude Opus 4.8 claude-opus-4-8 AnthropicProprietary	59coding index	1,540Code Arena rank #23	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#24	Claude Opus 4.6 claude-opus-4-6 AnthropicProprietary	58coding index	1,537Code Arena rank #27	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#25	Gemini 3.6 Flash High gemini-3.6-flash-high GoogleProprietary	56coding index	1,538Code Arena rank #25	Not listedNo confident match	161 tok/soutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#26	Claude Sonnet 4.6 claude-sonnet-4-6 AnthropicProprietary	54coding index	1,522Code Arena rank #30	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#27	Seed 2.1 Pro Preview seed-2.1-pro-preview ByteDanceProprietary	54coding index	1,522Code Arena rank #29	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-bench
#28	Grok 4.5 grok-4.5 xAIProprietary	53coding index	1,555Code Arena rank #21	Not listedNo confident match	48 tok/soutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis
#29	Qwen3.7 Max 20260517 qwen3.7-max-20260517 AlibabaProprietary	53coding index	1,517Code Arena rank #33	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-bench
#30	Hy3 hy3 TencentOpen weights	51coding index	1,511Code Arena rank #34	Not listedNo confident match	Not listedoutput speed	PlannedTerminal-Bench + Aider feed	Code ArenaSWE-benchArtificial Analysis

Qwen3.8 Max 0902

qwen3.8-max-0902

AlibabaProprietary

100coding index

Code Arena

1,691Code Arena rank #1

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-bench

Claude Opus 5 Max

claude-opus-5-max

AnthropicProprietary

99coding index

Code Arena

1,688Code Arena rank #2

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-bench

Kimi K3 Max

kimi-k3-max

Moonshot AIProprietary

95coding index

Code Arena

1,674Code Arena rank #3

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-bench

Hy4 Preview

hy4-preview

TencentOpen weights

83coding index

Code Arena

1,629Code Arena rank #6

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-bench

Qwen3.8 Max

qwen3.8-max

AlibabaProprietary

78coding index

Code Arena

1,669Code Arena rank #4

SWE-bench

Not listedNo confident match

Speed

40 tok/soutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

GLM 5.3 Max

glm-5.3-max

Z.aiOpen weights

77coding index

Code Arena

1,608Code Arena rank #11

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-bench

Claude Opus 5 High

claude-opus-5-high

AnthropicProprietary

77coding index

Code Arena

1,661Code Arena rank #5

SWE-bench

Not listedNo confident match

Speed

47 tok/soutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

Gemini 3.7 Flash High

gemini-3.7-flash-high

GoogleProprietary

74coding index

Code Arena

1,587Code Arena rank #14

SWE-bench

Not listedNo confident match

Speed

286 tok/soutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

GLM 5.2 Max

glm-5.2-max

Z.aiOpen weights

71coding index

Code Arena

1,585Code Arena rank #15

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-bench

#10

DeepSeek V4 Pro High 20260813

deepseek-v4-pro-high-20260813

DeepSeekOpen weights

71coding index

Code Arena

1,584Code Arena rank #16

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-bench

#11

DeepSeek V4 Flash High

deepseek-v4-flash-high

DeepSeekOpen weights

70coding index

Code Arena

1,581Code Arena rank #17

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-bench

#12

Claude Fable 5

claude-fable-5

AnthropicProprietary

70coding index

Code Arena

1,628Code Arena rank #8

SWE-bench

Not listedNo confident match

Speed

59 tok/soutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

#13

Qwen3.8 Flash Next

qwen3.8-flash-next

AlibabaProprietary

70coding index

Code Arena

1,620Code Arena rank #9

SWE-bench

Not listedNo confident match

Speed

84 tok/soutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

#14

Grok 4.6 High

grok-4.6-high

xAIProprietary

70coding index

Code Arena

1,629Code Arena rank #7

SWE-bench

Not listedNo confident match

Speed

50 tok/soutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

#15

GPT-5.6 Sol xHigh

gpt-5.6-sol-xhigh

OpenAIProprietary

68coding index

Code Arena

1,616Code Arena rank #10

SWE-bench

Not listedNo confident match

Speed

76 tok/soutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

#16

Claude Opus 4.8 High

claude-opus-4-8-high

AnthropicProprietary

65coding index

Code Arena

1,563Code Arena rank #18

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-bench

#17

Claude Opus 4.7

claude-opus-4-7

AnthropicProprietary

64coding index

Code Arena

1,557Code Arena rank #19

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

#18

GLM 5.3 Flash

glm-5.3-flash

Z.aiOpen weights

64coding index

Code Arena

1,604Code Arena rank #12

SWE-bench

Not listedNo confident match

Speed

42 tok/soutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

#19

Claude Opus 4.7 High

claude-opus-4-7-high

AnthropicProprietary

63coding index

Code Arena

1,556Code Arena rank #20

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-bench

#20

Qwen3.8 27b

qwen3.8-27b

AlibabaOpen weights

63coding index

Code Arena

1,599Code Arena rank #13

SWE-bench

Not listedNo confident match

Speed

49 tok/soutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

#21

Claude Opus 4.6 High

claude-opus-4-6-high

AnthropicProprietary

61coding index

Code Arena

1,546Code Arena rank #22

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-bench

#22

Muse Spark 1.1

muse-spark-1.1

MetaProprietary

59coding index

Code Arena

1,540Code Arena rank #24

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

#23

Claude Opus 4.8

claude-opus-4-8

AnthropicProprietary

59coding index

Code Arena

1,540Code Arena rank #23

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

#24

Claude Opus 4.6

claude-opus-4-6

AnthropicProprietary

58coding index

Code Arena

1,537Code Arena rank #27

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

#25

Gemini 3.6 Flash High

gemini-3.6-flash-high

GoogleProprietary

56coding index

Code Arena

1,538Code Arena rank #25

SWE-bench

Not listedNo confident match

Speed

161 tok/soutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

#26

Claude Sonnet 4.6

claude-sonnet-4-6

AnthropicProprietary

54coding index

Code Arena

1,522Code Arena rank #30

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

#27

Seed 2.1 Pro Preview

seed-2.1-pro-preview

ByteDanceProprietary

54coding index

Code Arena

1,522Code Arena rank #29

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-bench

#28

Grok 4.5

grok-4.5

xAIProprietary

53coding index

Code Arena

1,555Code Arena rank #21

SWE-bench

Not listedNo confident match

Speed

48 tok/soutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

#29

Qwen3.7 Max 20260517

qwen3.7-max-20260517

AlibabaProprietary

53coding index

Code Arena

1,517Code Arena rank #33

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-bench

#30

Hy3

hy3

TencentOpen weights

51coding index

Code Arena

1,511Code Arena rank #34

SWE-bench

Not listedNo confident match

Speed

Not listedoutput speed

Run mode

PlannedTerminal-Bench + Aider feed

Code ArenaSWE-benchArtificial Analysis

Benchmark guide

What the scores mean.

A quick reading key for comparing coding models without confusing source coverage, preference scores, and real issue-resolution benchmarks.

Higher: coding indexMissing rows are not zero

What does the coding index measure?

The coding index combines Code Arena and SWE-bench signals where available, with output speed used as a small tie-breaker. It is a practical coding comparison, not a guarantee that the model will solve every repository task.

How should I read Code Arena versus SWE-bench?

Code Arena is a public preference-style benchmark for coding outputs, while SWE-bench measures real software issue resolution. Code Arena is broader and more available; SWE-bench is more task-specific and harder to match across model names.

Why does Terminal-Bench show as planned?

Terminal-Bench is relevant for agentic coding workflows, but this page does not mix it into rankings until a reliable source feed is wired. Planned benchmarks are called out so the table does not imply hidden or invented scores.

Why are some SWE-bench cells empty?

Empty SWE-bench cells mean there is no public model row or no confident match to the model name in this table. They should be read as missing source coverage, not as a zero percent result.