What is the best agentic AI model?

The top of this table shows the current leader on our agentic composite. Terminal-Bench and SWE-bench Verified dominate the weighting because they measure real multi-step task completion, not one-shot answers.

Are these live scores?

Yes — the underlying benchmark data is refreshed every couple of hours from the public sources listed at the bottom of the page.

Agentic leaderboard

Best agentic LLMs

Agentic workloads reward long-horizon planning, tool-use and recovery from errors. This ranking weights Terminal-Bench (40%), SWE-bench Verified (30%), Aider Polyglot (15%), LiveBench (10%) and AA Index (5%).

Composite weightingHow this is calculated →

Terminal-Bench40%
Terminal-Bench
SWE-bench30%
SWE-bench Verified
Aider Polyglot15%
Aider Polyglot
LiveBench10%
LiveBench
AA Index5%
Artificial Analysis Intelligence Index

Category leader

#01Anthropic

Anthropic: Claude Fable 5.

Claude Fable 5 is a Mythos-class model from Anthropic, built for autonomous knowledge work and coding. It supports text, image, and file inputs with text output, with reasoning support and...

Composite score

73.9

/ 100

+1.9 vs #2

Ranking

N=20

#	Model	Composite	Terminal-Bench	SWE-bench	Aider Polyglot	LiveBench	AA Index
01	Anthropic: Claude Fable 5Anthropic	73.9	59.8	82.1	88.4	84.6	72
02	MoonshotAI: Kimi K3Moonshot	71.9	57.2	80.3	86.9	83.8	71
03	xAI: Grok 4.5xAI	70.7	56.1	78.8	85.6	82.9	70
04	DeepSeek: DeepSeek V4 ProDeepSeek	69.8	55.5	77.4	84.8	81.9	69
05	OpenAI: GPT-5.6 SolOpenAI	68.7	58.2	72.4	78.1	82.4	74
06	Qwen: Qwen3.7 MaxAlibaba	68.3	53.9	75.8	83.7	80.8	68
07	Anthropic: Claude Opus 4.8Anthropic	67.3	52.7	74.9	82.8	80.1	67
08	Anthropic: Claude Opus 4.8 (Fast)Anthropic	67.3	52.7	74.9	82.8	80.1	67
09	Z.ai: GLM 5.2Z.ai	66.3	51.8	73.6	81.9	79.4	66
10	NVIDIA: Nemotron 3 UltraNVIDIA	65.1	50.6	72.2	80.8	78.6	65
11	Sakana: Fugu UltraSakana	64.1	49.7	70.8	79.9	77.9	64
12	OpenAI: GPT-5.6 Terra ProOpenAI	63.6	49.3	70.4	79.4	77.5	63
13	OpenAI: GPT-5.6 TerraOpenAI	63.3	48.9	69.9	79.1	77.2	63
14	MoonshotAI: Kimi K2.7 CodeMoonshot	62.5	47.4	70.1	79.6	75.3	60
15	Meta: Muse Spark 1.1Meta	62.3	47.9	68.8	78.6	76.5	62
16	MiniMax: MiniMax M3MiniMax	61.3	46.8	67.6	77.8	75.8	61
17	Z.ai: GLM 5.1Z.ai	60.4	45.9	66.9	77.1	74.5	59
18	xAI: Grok 4.3xAI	59.6	45.2	65.8	76.4	73.9	58
19	MoonshotAI: Kimi K2.6Moonshot	59.0	44.6	65.2	75.9	73.2	57
20	Mistral: Mistral Medium 3.5Mistral	57.7	43.4	63.7	74.8	72.3	56

Frequently asked

What is the best agentic AI model?: The top of this table shows the current leader on our agentic composite. Terminal-Bench and SWE-bench Verified dominate the weighting because they measure real multi-step task completion, not one-shot answers.
Are these live scores?: Yes — the underlying benchmark data is refreshed every couple of hours from the public sources listed at the bottom of the page.