GPT-6.1 Terra added
The newest model in the AIM LLM Index is GPT-6.1 Terra.
AIM LLM Index
v1.10One transparent index combining 5 public benchmarks with AIMultiple's 4 agentic, RAG and enterprise-reasoning evaluations. See AIM LLM Index Methodology for further details.
TOP MODEL
Claude Opus 5.5
Index 78
BEST VALUE
GPT-6.1 Terra
$2.00 / 1M
FASTEST MODEL
Muse Spark Open
0.31s TTFT
Cost per Task vs Score
Average cost of one task across benchmarks, against the LLM Index
*Each model's cost per task averaged with equal weight over 7 benchmarks (all but MMMU-Pro, which covers too few models). Only models with a cost on every one of them are shown. ~ marks an average that includes an estimated cost.
AIM LLM Index
Leaderboard
The highest-scoring models across all benchmarks.
Filter & Sort
# | Model | Index | AIM-Agentic-RAG | AIM-AI-Bias | AIM-FinanceReasoning | AIM-Text-to-SQL | CritPt | FrontierMath | MMMU-Pro | SciCode | Terminal-Bench 4.0 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 |  Claude Opus 5.5 Anthropic | 78 | 90 | 94 | 92 | 88 | 32 | 95 | 88 | 67 | 60 |
| 2 |  GPT-6 Astra OpenAI | 74 | 84 | 98 | 87 | 66 | 32 | 98 | 87 | 56 | 60 |
| 3 | GPT-6.1 Sol OpenAI | 74 | 81 | 98 | 90 | 65 | 32 | 100 | 86 | 56 | 56 |
| 4 | Claude Sonnet 5.5 Anthropic | 74 | 85 | 93 | 89 | 74 | 31 | 80 | 86 | 61 | 64 |
| 5 | Claude Fable 5.1 Anthropic | 74 | 84 | 98 | 90 | 67 | 31 | 88 | 86 | 63 | 55 |
| 6 |  Gemini 4 Argon Google | 73 | 94 | 98 | 91 | 79 | 27 | 78 | - | 62 | 57 |
| 7 | GPT-6.1 Terra OpenAI | 71 | 84 | 95 | 87 | 57 | 28 | 93 | 85 | 55 | 56 |
| 8 |  Muse Wonder Meta | 70 | 86 | 95 | 92 | 56 | 28 | 71 | 85 | 60 | 52 |
| 9 | GPT-6 Sol OpenAI | 70 | 82 | 96 | 89 | 54 | 31 | 90 | 83 | 58 | 44 |
| 10 |  Kimi K3.1 Pro Moonshot AI | 69 | 85 | 91 | 90 | 52 | 25 | - | 84 | 61 | 67 |
Page 1 of 3
Cost$8.00
Latency3.15s
Context1M
TTFT3.15s
AIM-Agentic-RAG
90
AIM-AI-Bias
94
AIM-FinanceReasoning
92
AIM-Text-to-SQL
88
CritPt
32
FrontierMath
95
MMMU-Pro
88
SciCode
67
Terminal-Bench 4.0
60
Cost$20.00
Latency3.90s
Context1M
TTFT3.90s
AIM-Agentic-RAG
84
AIM-AI-Bias
98
AIM-FinanceReasoning
87
AIM-Text-to-SQL
66
CritPt
32
FrontierMath
98
MMMU-Pro
87
SciCode
56
Terminal-Bench 4.0
60
Cost$4.00
Latency3.50s
Context1M
TTFT3.50s
AIM-Agentic-RAG
81
AIM-AI-Bias
98
AIM-FinanceReasoning
90
AIM-Text-to-SQL
65
CritPt
32
FrontierMath
100
MMMU-Pro
86
SciCode
56
Terminal-Bench 4.0
56
Cost$4.00
Latency1.27s
Context1M
TTFT1.27s
AIM-Agentic-RAG
85
AIM-AI-Bias
93
AIM-FinanceReasoning
89
AIM-Text-to-SQL
74
CritPt
31
FrontierMath
80
MMMU-Pro
86
SciCode
61
Terminal-Bench 4.0
64
Cost$20.00
Latency3.33s
Context1M
TTFT3.33s
AIM-Agentic-RAG
84
AIM-AI-Bias
98
AIM-FinanceReasoning
90
AIM-Text-to-SQL
67
CritPt
31
FrontierMath
88
MMMU-Pro
86
SciCode
63
Terminal-Bench 4.0
55
Cost$4.00
Latency4.98s
Context1M
TTFT4.98s
AIM-Agentic-RAG
94
AIM-AI-Bias
98
AIM-FinanceReasoning
91
AIM-Text-to-SQL
79
CritPt
27
FrontierMath
78
SciCode
62
Terminal-Bench 4.0
57
Cost$2.00
Latency2.02s
Context1M
TTFT2.02s
AIM-Agentic-RAG
84
AIM-AI-Bias
95
AIM-FinanceReasoning
87
AIM-Text-to-SQL
57
CritPt
28
FrontierMath
93
MMMU-Pro
85
SciCode
55
Terminal-Bench 4.0
56
Cost$8.00
Latency35.22s
Context1M
TTFT35.22s
AIM-Agentic-RAG
86
AIM-AI-Bias
95
AIM-FinanceReasoning
92
AIM-Text-to-SQL
56
CritPt
28
FrontierMath
71
MMMU-Pro
85
SciCode
60
Terminal-Bench 4.0
52
Cost$4.00
Latency2.46s
Context1M
TTFT2.46s
AIM-Agentic-RAG
82
AIM-AI-Bias
96
AIM-FinanceReasoning
89
AIM-Text-to-SQL
54
CritPt
31
FrontierMath
90
MMMU-Pro
83
SciCode
58
Terminal-Bench 4.0
44
Cost$4.00
Latency49.60s
Context1M
TTFT49.60s
AIM-Agentic-RAG
85
AIM-AI-Bias
91
AIM-FinanceReasoning
90
AIM-Text-to-SQL
52
CritPt
25
MMMU-Pro
84
SciCode
61
Terminal-Bench 4.0
67
Page 1 of 3
Frontier Over Time
LLM Index by model release date
Model × Benchmark
Top models on selected benchmarks
Cost vs Performance
Blended cost against LLM Index
AIM LLM Index Methodology
A model's index score is the average of its results across the benchmarks below. Scores already on a 0-100 scale are used as-is; others are rescaled to 0-100. A model missing a benchmark has it excluded rather than scored as zero, and every benchmark carries equal weight. A model needs to be scored in at least 2 AIMultiple benchmarks to be ranked. MMMU-Pro results come from 2 separate sources.
Benchmarks We Used
AIMultiple
AIM-Agentic-RAGMulti-database routing & SQL query generation across 11 databases (BIRD-SQL, 759 questions).AIMultiple
AIM-AI-BiasMultiple-choice bias evaluation: 64 questions across 12 bias categories, each asked 5 times in two runs (v1 and v2), across 109 models. The score is the share of answers that pick the unbiased option, averaged over the two runs.AIMultiple
AIM-FinanceReasoningComplex multi-step financial reasoning across 238 hard FinanceReasoning questions.AIMultiple
AIM-Text-to-SQLNatural-language-to-SQL query generation accuracy across 35+ LLMs.Public
CritPtResearch-level physics problems requiring multi-step derivation.Public
FrontierMathUnpublished, expert-authored research math problems (Tiers 1–4).Public
MMMU-ProCollege-level multimodal understanding across disciplines, hardened against text-only shortcuts.Public
SciCodeResearch-level scientific coding across physics, math, biology & chemistry (338 subproblems)Public
Terminal-Bench 4.0The current Terminal-Bench generation: agentic terminal tasks scored by resolution rate.Download all data
Raw results of 73.9 thousand data points from the AIMultiple benchmarks this index is built from. Download them as a ZIP file containing 4 CSV files and a README.
Last updated: October 30, 2026
DownloadWant the granular data behind it? Join Premium
Changelog
62 updates- v1.10
2 models entered the AIM LLM Index: DeepSeek V4.1 Flash 1017, Muse Spark Open.
- v1.10
Added 113 models to the dataset: Claude 3 Haiku, Claude Opus 4.7 (max), Command A, DeepSeek R1 (Jan), DeepSeek R1 0528 Qwen3 8B and 108 more.
- v1.10
Added Qwen3.6 35B A3B to the dataset.
- v1.10
3 models entered the AIM LLM Index: GPT-6.1 Terra, Kimi K3.1 Pro, MiniMax M3.1 Flash.
- v1.10
Added 3 models to the dataset: GPT-6.1 Terra, Kimi K3.1 Pro, MiniMax M3.1 Flash.
- v1.10
Added 2 models to the dataset: DeepSeek V4.1 Flash 1017, Muse Spark Open.
- v1.10
Muse Wonder entered the AIM LLM Index.
- v1.10
Added Muse Wonder to the dataset.
- v1.10
Gemini 4 Argon entered the AIM LLM Index.
- v1.10
Added Gemini 4 Argon to the dataset.
LLM Release TimelineThe pace of AI innovation
AI Labs
Gemini 3.8 Flash Lite TTS +1 moreGemini 3.8 Flash Lite TTSGemini 3.8 Flash TTS
23 SepGemini 3.5 Transcribe
25 SepNano Banana 2.1
06 OctGemini 4 Argon
13 OctGemini Omni 1.1 Flash
04 NovGemini Omni 1.2 Flash
05 NovGPT Image 2.5 Flare +1 moreGPT Image 2.5 FlareGPT Image 2.5 Sunburst
09 SepGPT-6 Luna +1 moreGPT-6 LunaGPT-6 Sol
22 SepGPT-6.1 Sol
29 SepGPT-6 Luna Decisions
06 OctGPT-6.1 Terra
26 OctGPT Image 2.6 Flare +1 moreGPT Image 2.6 FlareGPT Image 2.6 Sunburst
04 Nov
GLM 5.3 FlashX
18 SepGLM 5.3 Prime
23 SepGLM 5.3 FlashX Legal
02 NovKimi K3.1 Pro
23 Oct
MiniMax M3.1 Flash
19 Oct
DeepSeek V4.1 Flash
10 SepDeepSeek V4.1 Flash 1017
17 Oct
Qwen3.8 Omni Flash
21 SepQwen3.8 Max Prime
23 SepQwen Image 2.1 Pro +1 moreQwen Image 2.1 ProQwen Image 2.1 Turbo
16 OctMuse Voice Transcribe 1.0
11 SepMuse Wonder
12 OctMuse Spark Open
15 Oct
Solar Mini 4
23 SepSolar Decide
28 SepSolar Decide Flash
08 OctClaude Opus 5.5
22 SepClaude Sonnet 5.5
28 SepClaude Haiku 5.5
07 Oct
Grok 4.7
21 SepGrok Imagine Video 1.5 Lite
06 Oct
MiMo-V2.6-Flash +2 moreMiMo-V2.6-FlashMiMo-V2.6-ProMiMo-V2.6-Pro-UltraSpeed
21 SepExplore LLM Use Cases, Analyses & Benchmarks
- LLMBenchmarkOct 6
Large Language Models in Cybersecurity
We evaluated 7 large language models across 9 cybersecurity domains using SecBench, a large-scale and multi-format benchmark for security tasks. We tested each model on 44,823 multiple-choice questions (MCQs) and 3,087 short-answer questions (SAQs), covering data security, identity & access management, network security, vulnerability management, and cloud security. MCQs (Multiple-Choice Questions) benchmarking: SAQs (Short Answer…
- LLMInsightSep 21
LLM VRAM Calculator for Self-Hosting
Self-hosting an LLM means running inference on hardware the operator controls rather than via a third-party API, which changes the cost, data control, and privacy profile. Whether a model runs at all depends on memory. The calculator estimates the VRAM or unified memory a model needs to run locally, based on the model, its precision,…
- LLMBenchmarkSep 18
Agentic IT: Can AI Agents Design a Benchmark
We tested 16 models on a benchmark design in text-to-SQL and tool calling. Each model built one benchmark per topic, for a total of 32 submissions. No submission passed every rubric criterion. The agents could build and run tests, but none demonstrated both a blank-answer test and a correct-answer test of their own scorer. Text-to-SQL…
- LLMBenchmarkSep 17
AIM Enterprise: Agentic Enterprise Benchmark
Enterprises use LLMs every day for their regular tasks. To find the most cost-efficient LLMs, we designed AIM Enterprise, an agentic enterprise benchmark, where we used 69 real enterprise tasks across strategy, marketing, HR, sales, and operations. Two judge models scored every file that passed the checks. On 32.7% of the individual scores the two…
- LLMInsightSep 15
LLM Scaling Laws: Analysis from AI Researchers
Large language models predict the next token based on patterns learned from text data. The term LLM scaling laws refers to empirical regularities that link model performance to the amount of compute, training data, and model parameters used during training. To understand how these relationships influence modern model design in practice, we reviewed findings from…
- LLMBenchmarkSep 15
Audience Simulation: Can LLMs Predict Human Behavior?
In marketing, evaluating how accurately LLMs predict human behavior is crucial for assessing their effectiveness in anticipating audience needs and recognizing the risks of misalignment, ineffective communication, or unintended influence. Audience simulation with LLMs enables the modeling of virtual audiences, helping organizations anticipate reactions to content or products without relying on costly surveys or focus…
- LLMInsightSep 2
LLM Observability Tools: Weights & Biases, Langsmith
LLM applications have expanded from single-turn chats into multi-step agents that use tools, query databases, and coordinate with other models, making their behavior harder to interpret. LLM observability provides continuous visibility into these complex workflows, helping organizations monitor quality, detect failures, troubleshoot issues, and manage performance and costs. W&B Weave is Weights & Biases‘ LLM…
- LLMInsightSep 1
LLM Fine-Tuning Guide for Enterprises
Follow the links for the specific solutions to your LLM output challenges. If your LLM: The widespread adoption of large language models (LLMs) has improved our ability to process human language. However, their generic training often results in suboptimal performance for specific tasks. To overcome this limitation, fine-tuning methods are employed to tailor LLMs to…
Filters