/ THE FRONTIER MODELS INDEX™ · VOLUME I · Q2 2026

Which AI model should you actually use?

A rating system for every major frontier AI model. 59 models rated across six capability dimensions — General Reasoning, Code Generation, Math & STEM, Tool Use & Agency, Multimodal, and Safety & Alignment. Sourced from public benchmarks (SWE-Bench, GPQA-Diamond, AIME, MMLU-Pro, TAU-bench, HarmBench), red-team disclosures, and FASCIA's proprietary evaluation suite. Updated quarterly.

59 models rated 6 capability dimensions 30+ benchmark sources Q2 2026 edition
Grade
Model
Score
Capability
A
Claude Opus 4.6
Frontier · Anthropic · Anthropic · Apr 2026
94/100
Excellent
→
A
Claude Sonnet 4.6
Frontier · Anthropic · Anthropic · Mar 2026
92/100
Excellent
→
A-
Claude Haiku 4.5
Frontier · Anthropic · Anthropic · Q1 2026
89/100
Excellent
→
A
GPT-5
Frontier · OpenAI · OpenAI · Sep 2025
92/100
Excellent
→
A-
GPT-5 mini
Frontier · OpenAI · OpenAI · Sep 2025
88/100
Excellent
→
A
OpenAI o3
Reasoning · OpenAI · OpenAI · Jan 2025
91/100
Excellent
→
B+
OpenAI o3-mini
Reasoning · OpenAI · OpenAI · Jan 2025
86/100
Good
→
A-
OpenAI o4-mini
Reasoning · OpenAI · OpenAI · Apr 2025
88/100
Excellent
→
A-
Gemini 2.5 Pro
Frontier · Google · Google DeepMind · Q1 2026
90/100
Excellent
→
B+
Gemini 2.5 Flash
Frontier · Google · Google DeepMind · Q1 2026
85/100
Good
→
B+
Grok 4
Frontier · xAI · xAI · Jul 2025
85/100
Good
→
B
Grok 3
Frontier · xAI · xAI · Feb 2025
80/100
Good
→
A-
Llama 4 Behemoth
Open-Weights · Meta · Meta AI · Apr 2025
88/100
Excellent
→
B+
Llama 4 Maverick
Open-Weights · Meta · Meta AI · Apr 2025
83/100
Good
→
B
Llama 4 Scout
Open-Weights · Meta · Meta AI · Apr 2025
80/100
Good
→
B
Llama 3.3 70B
Open-Weights · Meta · Meta AI · Dec 2024
79/100
Good
→
B-
Llama 3.1 405B
Open-Weights · Meta · Meta AI · Jul 2024
76/100
Good
→
B+
Mistral Large 2
Frontier · Mistral · Mistral AI · Jul 2024
84/100
Good
→
B
Mistral Medium 3
Frontier · Mistral · Mistral AI · May 2025
80/100
Good
→
B-
Mixtral 8x22B
Open-Weights · Mistral · Mistral AI · Apr 2024
75/100
Good
→
A-
DeepSeek R1
Reasoning · DeepSeek · DeepSeek · Jan 2025
89/100
Excellent
→
B+
DeepSeek V3
Open-Weights · DeepSeek · DeepSeek · Dec 2024
85/100
Good
→
A-
DeepSeek V3.5
Open-Weights · DeepSeek · DeepSeek · Q1 2026
87/100
Excellent
→
B+
Qwen 3
Open-Weights · Alibaba · Alibaba · Q1 2026
85/100
Good
→
B+
Qwen 2.5 Coder
Specialty · Code · Alibaba · Q3 2024
83/100
Good
→
B
Command R+
Enterprise · Cohere · Cohere · Apr 2024
81/100
Good
→
B-
Command R
Enterprise · Cohere · Cohere · Mar 2024
76/100
Good
→
B-
Jamba 1.5 Large
Architecture · AI21 · AI21 Labs · Aug 2024
75/100
Good
→
B
Phi-4
Small Model · Microsoft · Microsoft Research · Dec 2024
80/100
Good
→
B-
Phi-3.5
Small Model · Microsoft · Microsoft Research · Aug 2024
74/100
Good
→
B
Nova Pro
Frontier · Amazon · Amazon · Dec 2024
79/100
Good
→
B-
Nova Lite
Frontier · Amazon · Amazon · Dec 2024
74/100
Good
→
B-
Reka Core
Frontier · Reka · Reka AI · Apr 2024
77/100
Good
→
B+
Kimi K2
Frontier · Moonshot · Moonshot AI · Q2 2025
84/100
Good
→
B
Doubao Pro
Frontier · ByteDance · ByteDance · 2024 (multiple updates)
80/100
Good
→
B-
Yi-Large
Frontier · 01.AI · 01.AI (Kai-Fu Lee) · May 2024
76/100
Good
→
B+
GitHub Copilot (current generation)
Specialty · Code · GitHub / OpenAI · 2021+ continuous
83/100
Good
→
B+
Cursor (Composer model)
Specialty · Code · Anysphere · 2023+ continuous
85/100
Good
→
B
Replit Agent
Specialty · Code · Replit · 2024+
80/100
Good
→
A-
Midjourney V6 / V7
Multimodal · Image · Midjourney · V6 Dec 2023; V7 2025
89/100
Excellent
→
B+
DALL-E 3
Multimodal · Image · OpenAI · Oct 2023
83/100
Good
→
A-
FLUX.1
Multimodal · Image · Black Forest Labs · Aug 2024
87/100
Excellent
→
B
Stable Diffusion 3.5
Multimodal · Image · Stability AI · Oct 2024
80/100
Good
→
B+
Sora
Multimodal · Video · OpenAI · Dec 2024 (public)
85/100
Good
→
B+
Veo 3
Multimodal · Video · Google DeepMind · 2025
84/100
Good
→
B
Runway Gen-3 / Gen-4
Multimodal · Video · Runway · Gen-3 Jun 2024; Gen-4 2025
80/100
Good
→
A-
ElevenLabs (V3)
Multimodal · Audio / TTS · ElevenLabs · V3 2024
88/100
Excellent
→
B+
Suno
Multimodal · Music · Suno · V3 2024; V4 2024-2025
83/100
Good
→
B
Udio
Multimodal · Music · Uncharted Labs · Apr 2024
80/100
Good
→
F
Inflection 2.5 (Deprecated)
Deprecated · Inflection AI (Microsoft acquihire) · Mar 2024
28/100
Critical
→
D+
Stability AI base models (Pre-2025)
At-Risk · Stability AI · Multiple 2022-2024
48/100
High Risk
→
B
Perplexity Sonar
Specialty · Search · Perplexity · 2024 continuous
80/100
Good
→
B
Pixtral Large
Multimodal · Mistral · Mistral AI · Nov 2024
79/100
Good
→
B-
Hunyuan Large
Open-Weights · Tencent · Tencent · Nov 2024
75/100
Good
→
B-
Sailor 2
Open-Weights · Sea Group · Sea Group AI · 2024
74/100
Good
→
B
Aya Expanse
Open-Weights · Cohere · Cohere for AI · Oct 2024
78/100
Good
→
B
Gemma 3
Open-Weights · Google · Google DeepMind · Q1 2026
80/100
Good
→
B-
Falcon 3
Open-Weights · TII · Technology Innovation Institute (UAE) · Dec 2024
74/100
Good
→
B-
OLMo 2
Open-Weights · AI2 · Allen Institute for AI · Nov 2024
74/100
Good
→

The right model scales. The wrong one breaks.

Every AI lab claims their model is the best. The benchmarks tell a different story. The Frontier Models Index makes the actual capability visible — so engineering teams, procurement, and AI strategy leaders can pick models that ship, and avoid the ones that fail in production.

See the methodology Institutional API