llm leaderboard

Best LLM for agents

Agentic tool use covers whether a model drives tools to a finished outcome, stays steerable, and recovers when a command fails. Terminal-Bench and other evaluator-run agentic results identify the model together with its agent harness and effort; they are not pure model-only scores.

Aldena runs these models inside your team's projects. See what each one costs.

118 of 118 ranked models
Ranked models
rankmodelvendorcompositepricein / outbenchmarks%%scorescorescorescorescore
1
Claude Sonnet 5.5 (max)anthropic/claude-sonnet-5.5:max
Anthropic80.8$2.00 / $10.006 of 7 benchmarks63.9%16.013.819.50.815.0
2
Claude Opus 5.5 (max)anthropic/claude-opus-5.5:max
Anthropic79.6$4.00 / $20.005 of 7 benchmarks65.8%20.018.620.70.7
3
GPT-6 Astra (max)openai/gpt-6-astra:max
OpenAI78.0$10.00 / $50.006 of 7 benchmarks60.3%14.214.813.60.95.5
4
Claude Fable 5.1 (max)anthropic/claude-fable-5.1:max
Anthropic77.2$10.00 / $50.006 of 7 benchmarks57.3%14.315.713.10.811.6
5
GPT-6.1 Terra (max)openai/gpt-6.1-terra:max
OpenAI76.5$1.00 / $5.004 of 7 benchmarks12.115.810.60.9
6
GPT-6.1 Sol (max)openai/gpt-6.1-sol:max
OpenAI76.0$2.00 / $10.006 of 7 benchmarks60.9%12.513.311.20.94.5
7
Gemini 4 Carbon (high)google/gemini-4-carbon:high
Google73.7—4 of 7 benchmarks16.619.118.40.4
8
Claude Opus 5 (max)anthropic/claude-opus-5:max
Anthropic71.9$5.00 / $25.006 of 7 benchmarks52.1%7.54.910.60.913.6
9
Claude Fable 5 (high)anthropic/claude-fable-5:high
Anthropic71.0$10.00 / $50.005 of 7 benchmarks7.78.55.70.79.6
10
GPT-6 Sol (max)openai/gpt-6-sol:max
OpenAI69.9$2.00 / $10.006 of 7 benchmarks50.3%7.010.13.80.97.0
11
Kimi K3.1 Pro (max)moonshotai/kimi-k3.1-pro:max
MoonshotAI68.1$2.00 / $10.004 of 7 benchmarks5.27.52.80.9
12
Gemini 4 Argongoogle/gemini-4-argon
Google66.5$2.00 / $10.001 of 7 benchmarks90.7%
13
Claude Opus 5.5 (high)anthropic/claude-opus-5.5:high
Anthropic65.2$4.00 / $20.001 of 7 benchmarks13.6
14
Claude Opus 5 (xhigh)anthropic/claude-opus-5:xhigh
Anthropic64.7$5.00 / $25.002 of 7 benchmarks53.3%85.8%
15
Claude Fable 5.1anthropic/claude-fable-5.1
Anthropic64.7$10.00 / $50.001 of 7 benchmarks87.2%
16
Hy4tencent/hy4
Tencent63.2$0.417 / $1.25054 of 7 benchmarks4.75.08.10.1
17
Muse Spark 1.3 (max)meta/muse-spark-1.3:max
Meta63.0$1.25 / $4.255 of 7 benchmarks2.22.93.70.76.1
18
GPT-5.6 Sol (xhigh)openai/gpt-5.6-sol:xhigh
OpenAI62.7$4.00 / $20.005 of 7 benchmarks4.56.11.90.90.7
19
GPT-6 Astra (high)openai/gpt-6-astra:high
OpenAI62.6$10.00 / $50.001 of 7 benchmarks60.0%
20
Grok 4.7 (xhigh)x-ai/grok-4.7:xhigh
SpaceXAI62.6$2.00 / $6.006 of 7 benchmarks38.8%2.6-2.05.50.97.2
21
Claude Opus 5 (high)anthropic/claude-opus-5:high
Anthropic62.4$5.00 / $25.002 of 7 benchmarks49.7%12.6
22
Claude Opus 5.5anthropic/claude-opus-5.5
Anthropic62.4$4.00 / $20.001 of 7 benchmarks85.4%
23
GPT-6 Astra (xhigh)openai/gpt-6-astra:xhigh
OpenAI61.6$10.00 / $50.001 of 7 benchmarks59.7%
24
DeepSeek V4.1 Flash (max)deepseek/deepseek-v4.1-flash:max
DeepSeek61.5$0.051 / $0.5675 of 7 benchmarks2.13.52.50.37.9
25
Kimi K3 (max)moonshotai/kimi-k3:max
MoonshotAI61.5$0.192 / $16.006 of 7 benchmarks82.3%0.6-0.43.50.94.4
26
Qwen3.8 2.4T A95B (xhigh)qwen/qwen3.8-2.4t-a95b:xhigh
Qwen61.0$2.00 / $6.001 of 7 benchmarks84.5%
27
Claude Fable 5.1 (xhigh)anthropic/claude-fable-5.1:xhigh
Anthropic60.6$10.00 / $50.001 of 7 benchmarks57.6%
28
Claude Opus 4.8 (high)anthropic/claude-opus-4.8:high
Anthropic60.6$5.00 / $25.001 of 7 benchmarks7.2
29
Gemini 4 Argon (high)google/gemini-4-argon:high
Google60.6$2.00 / $10.005 of 7 benchmarks10.813.514.10.4-21.4
30
GLM 5.3z-ai/glm-5.3
Z.ai60.0$0.0699 / $1.401 of 7 benchmarks84.2%
31
Claude Sonnet 5 (high)anthropic/claude-sonnet-5:high
Anthropic59.3$2.00 / $10.005 of 7 benchmarks2.13.0-0.50.57.4
32
Gemini 3.5 Flash (high)google/gemini-3.5-flash:high
Google59.1$1.50 / $9.001 of 7 benchmarks83.6%
33
GPT-6 Astra (medium)openai/gpt-6-astra:medium
OpenAI58.7$10.00 / $50.001 of 7 benchmarks55.1%
34
Claude Fable 5anthropic/claude-fable-5
Anthropic58.2$10.00 / $50.001 of 7 benchmarks83.3%
35
Claude Fable 5.1 (high)anthropic/claude-fable-5.1:high
Anthropic57.7$10.00 / $50.001 of 7 benchmarks54.9%
36
GPT-5.5 (xhigh)openai/gpt-5.5:xhigh
OpenAI57.4$5.00 / $30.002 of 7 benchmarks75.3%9.6
37
Claude Fable 5.1 (medium)anthropic/claude-fable-5.1:medium
Anthropic56.7$10.00 / $50.001 of 7 benchmarks53.9%
38
MiMo-V2.6-Proxiaomi/mimo-v2.6-pro
Xiaomi56.2$0.435 / $0.875 of 7 benchmarks2.30.84.90.02.3
39
Muse Sparkmeta/muse-spark
Meta55.9—1 of 7 benchmarks82.2%
40
Hy4 previewtencent/hy4-preview
Tencent55.8$0.7506 / $2.25095 of 7 benchmarks0.4-0.44.2-0.06.5
41
Muse Wonder (max)meta/muse-wonder:max
Meta55.8$4.00 / $20.004 of 7 benchmarks1.92.65.4-0.8
42
GPT-6 Luna (max)openai/gpt-6-luna:max
OpenAI54.9$0.10 / $0.506 of 7 benchmarks15.2%1.53.8-0.60.95.2
43
GLM 5.2 (max)z-ai/glm-5.2:max
Z.ai54.5$0.006 / $6.805 of 7 benchmarks0.4-0.92.00.90.7
44
GPT-5.6 Solopenai/gpt-5.6-sol
OpenAI54.5$4.00 / $20.001 of 7 benchmarks81.8%
45
Grok 4.6 (xhigh)x-ai/grok-4.6:xhigh
SpaceXAI54.5$2.00 / $6.005 of 7 benchmarks-0.2-1.0-2.50.95.7
46
Grok 4.5x-ai/grok-4.5
SpaceXAI54.1$2.00 / $6.001 of 7 benchmarks5.1
47
GPT-6 Astra (low)openai/gpt-6-astra:low
OpenAI53.8$10.00 / $50.001 of 7 benchmarks50.9%
48
GLM 5.3 (max)z-ai/glm-5.3:max
Z.ai53.6$0.0699 / $1.406 of 7 benchmarks41.2%1.0-0.92.90.9-0.1
49
Inkling 2 Smallthinkingmachines/inkling-2-small
Thinking Machines53.2—5 of 7 benchmarks85.4%0.12.12.1-1.5
50
DeepSeek V4 Pro 0423 (high)deepseek/deepseek-v4-pro:high
DeepSeek52.8$0.8837 / $1.76751 of 7 benchmarks4.4
51
Claude Opus 4.7 (max)anthropic/claude-opus-4.7:max
Anthropic52.6$5.00 / $25.001 of 7 benchmarks79.1%
52
GPT-6 Luna (medium)openai/gpt-6-luna:medium
OpenAI51.8$0.10 / $0.504 of 7 benchmarks-2.00.1-6.30.9
53
Gemini 3.1 Pro Preview (high)google/gemini-3.1-pro-preview:high
Google51.7$2.00 / $12.001 of 7 benchmarks78.2%
54
GPT-5.5openai/gpt-5.5
OpenAI51.5$5.00 / $30.001 of 7 benchmarks4.1
55
Claude Opus 5 (medium)anthropic/claude-opus-5:medium
Anthropic50.8$5.00 / $25.001 of 7 benchmarks45.1%
56
GLM 5.2z-ai/glm-5.2
Z.ai50.8$0.006 / $6.801 of 7 benchmarks77.8%
57
Qwen3.8 Maxqwen/qwen3.8-max
Qwen50.1—5 of 7 benchmarks0.3-0.74.7-0.82.8
58
Claude Fable 5 (max)anthropic/claude-fable-5:max
Anthropic49.9$10.00 / $50.001 of 7 benchmarks44.9%
59
Claude Opus 4.6 (max)anthropic/claude-opus-4.6:max
Anthropic49.9$5.00 / $25.001 of 7 benchmarks76.8%
60
Inkling (xhigh)thinkingmachines/inkling:xhigh
Thinking Machines48.9$1.00 / $4.051 of 7 benchmarks76.0%
61
Claude Fable 5.1 (low)anthropic/claude-fable-5.1:low
Anthropic48.9$10.00 / $50.001 of 7 benchmarks43.3%
62
GPT-5.4 (high)openai/gpt-5.4:high
OpenAI48.9$2.50 / $15.001 of 7 benchmarks1.2
63
GPT-5.6 Terra (xhigh)openai/gpt-5.6-terra:xhigh
OpenAI48.8$2.00 / $12.005 of 7 benchmarks-1.90.2-7.50.90.9
64
Claude Opus 4.8 (max)anthropic/claude-opus-4.8:max
Anthropic48.5$5.00 / $25.002 of 7 benchmarks23.6%82.2%
65
GLM 5.1z-ai/glm-5.1
Z.ai48.0$1.40 / $4.401 of 7 benchmarks75.6%
66
Claude Haiku 5.5 (high)anthropic/claude-haiku-5.5:high
Anthropic47.8$0.10 / $0.504 of 7 benchmarks0.42.71.3-1.7
67
GPT-5.4 (xhigh)openai/gpt-5.4:xhigh
OpenAI46.2$2.50 / $15.001 of 7 benchmarks70.6%
68
Gemini 3.8 Flash (high)google/gemini-3.8-flash:high
Google46.1$0.75 / $3.756 of 7 benchmarks18.2%2.4-0.60.90.5-0.7
69
GPT-5.6 Sol (max)openai/gpt-5.6-sol:max
OpenAI45.9$4.00 / $20.001 of 7 benchmarks38.5%
70
GLM 5.3 Flashz-ai/glm-5.3-flash
Z.ai45.8$0.15 / $0.505 of 7 benchmarks-2.8-4.71.70.9-5.7
71
MiMo-V2.6-Flashxiaomi/mimo-v2.6-flash
Xiaomi45.7$0.14 / $0.285 of 7 benchmarks0.12.35.0-4.0-1.1
72
GPT-5.6 Luna (xhigh)openai/gpt-5.6-luna:xhigh
OpenAI45.6$0.20 / $1.201 of 7 benchmarks0.5
73
Gemini 3 Pro Previewgoogle/gemini-3-pro-preview
Google45.2—1 of 7 benchmarks70.3%
74
GLM 5.3 Flash (none)z-ai/glm-5.3-flash:none
Z.ai45.0$0.15 / $0.501 of 7 benchmarks35.5%
75
Claude Opus 4.5 (high)anthropic/claude-opus-4.5:high
Anthropic44.3$5.00 / $25.001 of 7 benchmarks69.8%
76
Claude Opus 5 (low)anthropic/claude-opus-5:low
Anthropic44.0$5.00 / $25.001 of 7 benchmarks34.9%
77
Claude Sonnet 4.6anthropic/claude-sonnet-4.6
Anthropic43.4$3.00 / $15.001 of 7 benchmarks69.5%
78
Muse Spark 1.1meta/muse-spark-1.1
Meta43.3$1.25 / $4.256 of 7 benchmarks88.1%-7.9-14.6-6.50.9-2.3
79
MiniMax M3.1 Flashminimax/minimax-m3.1-flash
MiniMax43.1$0.075 / $0.304 of 7 benchmarks-2.8-3.50.30.2
80
Qwen3.8 Max (0902) (max)qwen/qwen3.8-max-0902:max
Qwen43.0$2.00 / $6.001 of 7 benchmarks27.0%
81
GPT-6 Luna (low)openai/gpt-6-luna:low
OpenAI42.8$0.10 / $0.504 of 7 benchmarks-6.2-3.3-15.60.9
82
GPT-5.2 (xhigh)openai/gpt-5.2:xhigh
OpenAI42.4$1.75 / $14.001 of 7 benchmarks67.6%
83
Kimi K2p5moonshotai/kimi-k2p5
Moonshot AI41.5—1 of 7 benchmarks64.4%
84
GPT-5.6 Terra (max)openai/gpt-5.6-terra:max
OpenAI41.0$2.00 / $12.001 of 7 benchmarks21.5%
85
Nemotron 3 Ultra (thinking)nvidia/nemotron-3-ultra-550b-a55b:thinking
NVIDIA40.6$0.60 / $2.401 of 7 benchmarks63.1%
86
Muse Spark 1.2 (xhigh)meta/muse-spark-1.2:xhigh
Meta40.5$1.25 / $4.255 of 7 benchmarks-8.0-11.8-9.30.75.8
87
Grok 4.6 (high)x-ai/grok-4.6:high
SpaceXAI40.1$2.00 / $6.001 of 7 benchmarks19.7%
88
Gemini 3 Flash Previewgoogle/gemini-3-flash-preview
Google39.7$0.50 / $3.001 of 7 benchmarks62.0%
89
Claude Sonnet 4.5 (thinking)anthropic/claude-sonnet-4.5:thinking
Anthropic38.7$3.00 / $15.001 of 7 benchmarks59.5%
90
GPT-5.6 Luna (max)openai/gpt-5.6-luna:max
OpenAI38.1$0.20 / $1.201 of 7 benchmarks17.6%
91
Step 5 Previewstepfun/step-5-preview
StepFun38.0$1.00 / $2.705 of 7 benchmarks-3.4-1.1-0.7-1.60.2
92
Mistral Large 4mistralai/mistral-large-4-0
Mistral37.9$0.70 / $2.105 of 7 benchmarks-6.1-6.9-5.3-0.01.8
93
GLM 4.7z-ai/glm-4.7
Z.ai37.8$0.60 / $2.201 of 7 benchmarks58.1%
94
Gemini 3.1 Flash Lite (high)google/gemini-3.1-flash-lite:high
Google36.9$0.25 / $1.501 of 7 benchmarks57.1%
95
MiMo-V2.5-Proxiaomi/mimo-v2.5-pro
Xiaomi36.4$0.48 / $1.801 of 7 benchmarks-7.5
96
Qwen3.8 Flash Nextqwen/qwen3.8-flash-next
Qwen36.4—5 of 7 benchmarks-4.7-2.90.0-1.0-2.6
97
Muse Spark 1.3 (xhigh)meta/muse-spark-1.3:xhigh
Meta36.1$1.25 / $4.251 of 7 benchmarks14.8%
98
GPT-5.4 Mini (xhigh)openai/gpt-5.4-mini:xhigh
OpenAI36.0$0.75 / $4.501 of 7 benchmarks56.7%
99
Gemini 3.6 Flash (high)google/gemini-3.6-flash:high
Google35.8$0.75 / $3.751 of 7 benchmarks-18.2
100
Qwen3.8 27Bqwen/qwen3.8-27b
Qwen35.8$0.15 / $1.8755 of 7 benchmarks-4.6-5.4-2.90.1-5.9
101
GPT-5.1 (high)openai/gpt-5.1:high
OpenAI35.0$1.25 / $10.001 of 7 benchmarks50.1%
102
Gemini 3.7 Flash (high)google/gemini-3.7-flash:high
Google35.0$0.75 / $3.756 of 7 benchmarks12.4%-1.9-6.4-3.50.6-5.4
103
Grok 4.5 (high)x-ai/grok-4.5:high
SpaceXAI34.7$2.00 / $6.001 of 7 benchmarks12.4%
104
o3 Proopenai/o3-pro
OpenAI34.1$20.00 / $80.001 of 7 benchmarks44.5%
105
Claude Haiku 4.5anthropic/claude-haiku-4.5
Anthropic33.2$1.00 / $5.001 of 7 benchmarks40.2%
106
Claude Sonnet 5 (max)anthropic/claude-sonnet-5:max
Anthropic33.2$2.00 / $10.001 of 7 benchmarks12.1%
107
Qwen3.7 Maxqwen/qwen3.7-max
Qwen30.3$1.475 / $4.4255 of 7 benchmarks-7.1-11.8-10.7-0.6-1.6
108
Inkling Smallthinkingmachines/inkling-small
Thinking Machines30.0$0.45 / $1.206 of 7 benchmarks79.2%-17.7-26.7-27.3-3.65.4
109
Gemini 3.1 Pro Previewgoogle/gemini-3.1-pro-preview
Google29.3$2.00 / $12.005 of 7 benchmarks-4.3-7.7-11.5-0.2-23.3
110
Beam 1.0 501B A23Breflection/beam-1.0-501b-a23b
Reflection29.3$0.30 / $1.004 of 7 benchmarks-6.4-9.5-11.3-1.8
111
MiniMax M3minimax/minimax-m3
MiniMax29.1$0.30 / $1.205 of 7 benchmarks-10.4-10.3-17.7-1.10.6
112
Hy3tencent/hy3
Tencent26.5$0.0825 / $0.335 of 7 benchmarks-7.7-2.6-12.1-4.7-7.2
113
Qwen3.7 Plusqwen/qwen3.7-plus
Qwen25.7$0.32 / $1.285 of 7 benchmarks-7.9-11.3-10.4-1.9-6.3
114
Nemotron 3 Ultranvidia/nemotron-3-ultra-550b-a55b
NVIDIA24.0$0.60 / $2.404 of 7 benchmarks-14.4-15.4-22.2-1.4
115
Inklingthinkingmachines/inkling
Thinking Machines21.9$1.00 / $4.055 of 7 benchmarks-17.1-21.7-26.3-3.40.1
116
Nemotron 3.5 Super VLnvidia/nemotron-3.5-super-vl
NVIDIA19.7$0.15 / $0.474 of 7 benchmarks-16.3-17.8-18.9-7.0
117
Mistral Medium 3.5mistralai/mistral-medium-3-5
Mistral19.0$1.50 / $7.505 of 7 benchmarks-13.9-16.6-17.7-6.1-20.4
118
Solar Pro 4upstage/solar-pro4
Upstage15.8$0.15 / $0.605 of 7 benchmarks-22.8-20.1-31.3-3.7-25.6
How this ranks

Every benchmark value becomes a percentile among the models that have it, so accuracy scores, Elo ratings and word error rates compare without hand-tuned scaling. Metrics where lower is better are inverted first. Raw values are never summed or averaged across benchmarks. A model's mean percentile is then shrunk toward the mean of the models that were broadly benchmarked, so a model tested twice cannot outrank a broadly tested one on two lucky results. Turning a data source off runs that same ranking code again in your browser over the sources you left on.

A model scored on fewer than 3 of the 7 ranked benchmarks in this category still ranks here, on the benchmarks it does have, and its row carries a partial coverage mark. On an equal score it sits under the model that earned the same number across more of the board.

Data sources

Turn a source off to drop every benchmark it feeds and re-rank the board in your browser. Your choice follows you across the leaderboard pages.

  • LMArenaCC BY 4.0

    Arena ratings by LMArena, from the public leaderboard dataset.

  • MCP AtlasMIT

    All-1,000-task Pass Rate from Scale's current Performance Comparison, using the standardized MCP loop and 100-tool-call budget.

  • Terminal-Bench 4.1Apache-2.0

    Published Accuracy over 66 Terminal-Bench 4.1 tasks, run and verified by the benchmark team; results are model plus agent harness plus effort, not model-only evaluations.