NEWAdded Agnes 3.0 Flash & DeepSeek V4.1 Flash
Independent analysis of AI
Understand the frontier AI landscape to choose the optimal foundation model, pricing tier, and multi-cloud inference provider for your use case.
- Models ranked
- 772
- Categories
- 18
- Data sources
- 5
Highlights
01 · Real-Time Rankings
Top Ranked Models
Top 10 Models by Overall Podium Score
View All →| # | Model | Podium Score |
|---|---|---|
| 01 | Claude Mythos Preview Anthropic · Proprietary | 97.4 |
| 02 | Claude 5.5 Sonnet Anthropic · Proprietary | 94.8 |
| 03 | Kimi K3.1 Pro Moonshot AI · Open Weights | 94.2 |
| 4 | Eleven Multilingual v3 ElevenLabs · Proprietary | 94.0 |
| 5 | FLUX.2 Pro Black Forest Labs · Proprietary | 93.8 |
| 6 | Gemini 4 Argon Google · Proprietary | 93.6 |
| 7 | Claude Fable 5 Anthropic · Proprietary | 93.4 |
| 8 | GPT-Image 2 OpenAI · Proprietary | 93.1 |
| 9 | Grok Imagine Quality xAI · Proprietary | 92.6 |
| 10 | ElevenLabs Scribe v2 ElevenLabs · Proprietary | 92.5 |
02 · Evaluation Domains
Best by Use Case
Coding
Repository-level software engineering: SWE-Bench Verified and Pro, LiveCodeBench, SciCode, Terminal-Bench and HumanEval.
Leader: Claude Fable 5.1
Math
Mathematical problem solving from competition math to frontier research problems: AIME, MATH and FrontierMath.
Leader: Claude Mythos Preview
Reasoning
Deep reasoning and scientific knowledge: GPQA Diamond, Humanity’s Last Exam and ARC-AGI-2.
Leader: Claude Mythos Preview
Agentic
Autonomous tool use and computer operation: OSWorld, Toolathlon, MCP Atlas, τ²-Bench Retail and Apex Agents.
Leader: Kimi K3
Knowledge
Factual knowledge and hallucination resistance: SimpleQA, MMLU-Pro and Multilingual MMLU.
Leader: Claude Mythos Preview
Multimodal
Vision and document understanding: MMMU, MMMU-Pro, CharXiv Reasoning and ScreenSpot-Pro.
Leader: Claude Mythos Preview
03 · Research & Analysis
Latest Articles
04 · FAQ
Frequently Asked Questions
Every model gets a Podium Score (0–100): a weighted blend of four signals — 35% arena Elo from head-to-head human battles, 30% benchmark average, 20% Artificial Analysis Intelligence Index and 15% LLM Stats score. All signals are normalized to a common scale, so the score is a plausible middle ground between five independent leaderboards.
Claude Mythos Preview leads the podium with a score of 97.4. Claude 5.5 Sonnet (94.8) and Kimi K3.1 Pro (94.2) follow. Rankings shift weekly as new arena votes and benchmark results arrive.
We curate 772 flagship models across 18 categories (coding, math, reasoning, agentic, knowledge, multimodal, long context, speed, value and open weights), plus the full arena table with 726 models.
Five public leaderboards: Arena.ai, Artificial Analysis, LLM Stats, Vellum Leaderboard, LLMBase. Each source is re-synced weekly; see the Methodology page for the exact formula.