Quick takeaways
- The best model depends on your task, not the top of a benchmark table.
- Frontier models are close enough that cost, speed, and integration often decide the winner.
- Re-test quarterly because model releases change rankings fast.
How to think about AI model choice
Andrej Karpathy explains what large language models are good at and how to pick the right model for a task.
How this leaderboard works
This leaderboard ranks models by practical fit for common workflows. We weigh output quality, context window, speed, cost, and ecosystem. Benchmarks are a signal, but real-task performance and review effort matter more.
Each model gets an overall tier and a primary strength. Use the table to shortlist, then test on one of your own tasks before committing.
Leaderboard table
| Rank | Model | Maker | Best for | Context | Price / 1M tokens* | Key benchmark | Verdict |
|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 4.8 | Anthropic | Deep reasoning, coding, research | Very large | ~$90 | SWE-Bench, GPQA | Best for hard tasks |
| 2 | GPT-5.4 | OpenAI | General purpose, agents, multimodal | Very large | ~$120 | MMLU, MMMU | Best all-rounder |
| 3 | Gemini 3.1 Pro | Coding arena, long context | Largest | ~$110 | SWE-Bench, LiveBench | Best for long docs | |
| 4 | Claude Sonnet 5 | Anthropic | Daily writing, analysis, coding | Large | ~$18 | HumanEval, MMLU | Best daily driver |
| 5 | GPT-5.1 | OpenAI | Fast general tasks | Large | ~$12 | MMLU, HumanEval | Best fast option |
| 6 | Qwen 3.7 Max | Alibaba | Cost-efficient frontier work | Large | ~$3 | Multi-language, MMLU | Best value |
| 7 | DeepSeek-V3.5 | DeepSeek | Reasoning, coding, open weights | Large | ~$0.40 | HumanEval, MATH | Best open model |
| 8 | Grok 4 | xAI | Live search, social context | Very large | ~$15 | LiveBench, reasoning | Best real-time edge |
| 9 | Llama 4 | Meta | Self-hosting, customization | Large | Free weights + hosting | HumanEval, MMLU | Best for self-host |
| 10 | Claude Haiku 5 | Anthropic | High-volume simple queries | Large | ~$1.50 | Speed, cost | Best lightweight |
*Approximate blended input/output API cost per million tokens. Verify current pricing on provider sites. Use the LLM cost calculator to estimate your workload.
Winners by use case
Claude Sonnet 5
Careful tone, strong structure, and reliable editing for long-form and brand work.
Claude Opus 4.8
Best for complex code, architecture decisions, and repo-wide reasoning.
Gemini 3.1 Pro
Huge context window and strong long-document handling.
Qwen 3.7 Max
Frontier-level quality at a much lower price point.
Methodology
Rankings combine published benchmark signals, live API performance, pricing data, and hands-on testing across writing, coding, and research tasks. We update this page when major models are released or when pricing changes significantly.
We do not claim exact benchmark reproducibility. Use this leaderboard as a starting point, then validate on your own data.
Update log
| Date | Change |
|---|---|
| July 11, 2026 | Added price-per-1M-tokens estimates, benchmark references, and links to cost calculator and API pricing guide. |
| July 5, 2026 | Initial leaderboard published with workflow-fit tiers. |
Without AI vs. with AI
| Task | Without AI | With AI |
|---|---|---|
| Picking a model | Teams choose based on hype or a single headline score. | The leaderboard ranks models by workflow fit, cost, and context. |
| Tracking releases | Users rely on outdated recommendations. | The update log tracks when the leaderboard changes. |
| Comparing cost | Teams estimate budgets from scattered pricing pages. | Price-per-1M-token tiers give a quick cost comparison. |
| Finding use-case winners | Readers scan long tables for relevance. | Use-case cards highlight the best model for writing, coding, research, and budget. |
| Benchmark context | Headline scores drive decisions alone. | Methodology explains how benchmarks and hands-on testing combine. |
FAQ
How often is this leaderboard updated?
We update it quarterly or whenever a major model is released.
Why not just use the top benchmark model?
Benchmarks measure narrow skills. Your workflow may care more about speed, cost, or integration than a few percentage points on a test.
Are prices included?
We show price tiers rather than exact rates because pricing changes frequently. Check the provider's site for current API or subscription pricing.
Which model is best for beginners?
Start with Claude Sonnet 5 or GPT-5.1. Both are easy to use through familiar chat interfaces.
Should I use multiple models?
Yes, if different workflows have different needs. Keep the stack small enough to manage and review.
How is this leaderboard different from others?
It ranks by workflow fit, not just benchmark scores.
What is the best model overall?
Claude Opus 4.8 and GPT-5.4 are top all-rounders, but the best model depends on your task.
Are these prices exact?
No, they are approximate tiers. Check provider sites for current rates.
How do I test models myself?
Run a few real tasks from your workflow and score speed, quality, and review time.
What about open models?
DeepSeek, Qwen, and Llama are included and often win on cost and control.