New models: waiting for benchmarks
Released in the last 30 days and not yet scored by any benchmark we track, so they are not ranked yet. Each one moves into the table below by itself as soon as its first score is published.
Loading live benchmark data… |
Try these models
Cheapest current provider for the top models in the current view.
LLM benchmarks leaderboard: top models, November 2026
LLMs Tiger tracks 76 LLM benchmarks across 968 models; the 827 with enough results are combined into one aggregate score, so you can see at a glance which model leads on coding, reasoning, agents and chat quality, what it costs, and how old it is. Every number links back to the public source it came from. Scores last captured 2026-11-18.
| # | Model | Organisation | Aggregate score | Cheapest input $/M tokens |
|---|---|---|---|---|
| 1 | Claude Fable 5.2 (max) | Anthropic | 63.4 | $8.00 |
| 2 | Claude Opus 5.5 (max) | Anthropic | 63.1 | $4.00 |
| 3 | Gemini 4 Carbon (high) | Google DeepMind | 62.4 | $2.00 |
| 4 | GPT-6 Astra (max) | OpenAI | 62.3 | $10.00 |
| 5 | Claude Fable 5.1 (max) | Anthropic | 61.4 | $10.00 |
| 6 | Claude Sonnet 5.5 (max) | Anthropic | 60.7 | $2.00 |
| 7 | GPT-6.2 Astra (max) | OpenAI | 60.5 | $8.00 |
| 8 | GPT-6.1 Sol (max) | OpenAI | 60.4 | $2.00 |
| 9 | gemini-4-argon-high | 60.0 | $2.00 | |
| 10 | GPT-6.1 Terra (max) | OpenAI | 59.8 | $1.00 |
| 11 | Claude Opus 5 (max) | Anthropic | 59.1 | $5.00 |
| 12 | GPT-6 Sol (max) | OpenAI | 58.0 | $2.00 |
| 13 | GPT-5.6 Sol (max) | OpenAI | 58.0 | $2.00 |
| 14 | Claude Fable 5 (max) | Anthropic | 57.9 | $10.00 |
| 15 | claude-opus-4-8-xhigh-effort | Anthropic | 56.6 | $5.00 |
Best LLMs by task
- Best LLMs for Coding: Software-engineering benchmarks: resolving real GitHub issues (SWE-bench family), editing code with a tool (Aider), terminal and agentic coding tasks, competitive programming (ALE-Bench) and crowd-voted coding answers (LMArena Code).
- Best LLMs for Reasoning and Maths: Reasoning, maths and science benchmarks: graduate-level science questions (GPQA Diamond), competition and research mathematics (AIME-style problems, FrontierMath), expert exam questions (Humanity's Last Exam) and abstract puzzles (ARC-AGI-2).
- Best LLMs for Agents and Tool Use: Agent benchmarks: tool use over MCP servers, long professional tasks, computer use, game playing, research reports and paid freelance-style work.
- Best Open-Source LLMs: Every benchmark we track, restricted to models whose weights are public.
- Cheapest LLM APIs by Benchmark Score: The cheapest listed input price among the 25 highest-scoring models on the aggregate score.
LLM benchmarks tracked
GPQA Diamond · OTIS Mock AIME · Epoch ECI · MMLU-Pro · DTBench · CritPt · LMCA · SciCode · WeirdML · ARC-AGI-2 · ALE-Bench · Aider Code Editing · Chess Puzzles · SimpleQA Verified · FrontierMath T1-3 · SWE-bench Verified · Mystery Games · WebDev Arena · SimpleBench · FrontierMath T4 · Terminal-Bench 4.0 · Aider Polyglot · ProofBench · Fiction.liveBench · all 56 benchmark leaderboards →
New to LLM benchmarks?
A benchmark is a fixed test that every model takes under the same rules, which is what makes scores comparable. Our guide explains the main types, how to read a score and where benchmarks mislead: LLM benchmarks explained.
LLM benchmark FAQ
What are LLM benchmarks?
LLM benchmarks are standardised tests that score large language models on the same tasks, such as graduate-level science questions, competition maths, fixing real software bugs or acting as an agent. Because each model is given the same tasks and graded the same way, the scores let you compare models side by side (as long as the test setup matches).
Which LLM is best right now?
On our aggregate score, Claude Fable 5.2 (max) ranks first as of 2026-11-18. The best model depends on the task, so compare it on the benchmark closest to your use: coding, reasoning, agents or chat quality. The leaderboard lets you sort by any single benchmark.
How does LLMs Tiger rank models?
Each model gets an aggregate score built from its z-scores across every benchmark it has been measured on, shrunk toward the average when it has few results. 50 is the neutral midpoint, and higher is better. The method, the sources and the exclusions are on the methodology page.
Where do the benchmark scores come from?
From public, cited sources such as Epoch AI's Benchmarking Hub, LMArena, LiveBench, the SWE-bench and Aider leaderboards, ARC Prize and Hugging Face. Each score shows its source and capture date. We do not run our own tests.
How often is the data updated?
Scores are refreshed every 12 hours and API prices every hour. The newest score capture currently on the site is 2026-11-18.
Which strong LLM is cheapest to use?
Among the 25 highest-scoring models, DeepSeek V4.1 Flash 1017 (max) has the lowest listed input price at $0.036 per million tokens. Prices are the cheapest listing we found and change often.