The model cheatsheet

Which AI is best for which job?

Name: Which AI for which job — LLM model cheatsheet
Creator: MyTokenTracker
License: https://creativecommons.org/licenses/by/4.0/

There's no single "best" model — only the best one for the task in front of you. Here's what each is actually good at, ranked from real benchmark, human-preference, and price data. Not opinions.

See the cheatsheet But what does it cost? →

The cheatsheet

Best models for each job

Pick the job, get the shortlist. Each list is ranked by the metric that actually matters for that task — and refreshes as new models land.

🧠

Reasoning & hard problems

Deep multi-step thinking, math, analysis

1 Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) 60.1
2 GPT-5.5 (xhigh) 54.8
3 Gemini 3.5 Flash (high) 50.2
4 MO Kimi K3 (low) 46.6
5 MI MiniMax-M3 44.4

Ranked by Intelligence Index

💻

Writing & shipping code

Generating, refactoring, and fixing code

1 Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) 77.0
2 GPT-5.5 (xhigh) 74.9
3 MO Kimi K3 (low) 72.0
4 Gemini 3.5 Flash (high) 70.1
5 XI MiMo-V2.5-Pro 60.2

Ranked by Coding Index

🤖

Agents & tool use

Autonomous workflows that call tools

1 Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) 54.5
2 GPT-5.6 Luna (max) 45.6
3 Gemini 3.5 Flash (high) 37.4
4 MO Kimi K3 (low) 37.0
5 DeepSeek V4 Pro (Reasoning, Max Effort) 36.4

Ranked by Agentic Index

💬

General chat & writing

Everyday assistant, drafting, Q&A

1 claude-opus-4-6-thinking 1,503
2 gpt-5.4 1,499
3 gemini-3.6-flash 1,491
4 muse-spark-1.1 1,483
5 qwen3.7-max-preview 1,480

Ranked by LMArena (human votes)

⚡

Real-time & low latency

Voice, autocomplete, anything live

1 ST Step 3.7 Flash 413 tok/s
2 LA LFM2.5-VL-1.6B 401 tok/s
3 Gemini 3.5 Flash-Lite 368 tok/s
4 IB Granite 3.3 8B (Non-reasoning) 368 tok/s
5 gpt-oss-120b (low) 346 tok/s

Ranked by Output tokens/sec

📚

Huge documents & long context

Whole codebases, books, long transcripts

1 Llama 4 Scout 17b 128e Instruct Maas 10M
2 Gemini Exp 1206 2.1M
3 Grok 4 Fast Reasoning 2M
4 GPT 5.5 1.1M
5 DA Databricks Gemini 2 5 Flash 1M

Ranked by Context window

💎

Best bang for the buck

The most intelligence per dollar

1 Qwen3.5 4B (Non-reasoning) 266.7 pts/$
2 Gemma 4 E4B (Non-reasoning) 222.5 pts/$
3 DeepSeek V4 Flash (Reasoning, High Effort) 213.7 pts/$
4 XI MiMo-V2.5 212.6 pts/$
5 ST Step 3.5 Flash 2603 173.3 pts/$

Ranked by Intelligence per $/Mtok

🪙

High-volume on a budget

Cheap, good-enough, at scale

1 Llama 3.1 8b $0.035/Mtok
2 Meta Llama 3.2 1B Instruct $0.05/Mtok
3 CO Command R7b 12 2024 $0.066/Mtok
4 Mistral Small Latest $0.09/Mtok
5 ZA GLM 4 32b 0414 128k $0.1/Mtok

Ranked by Lowest blended $/Mtok

Ranked from live data · updated 8 hours ago. Model & provider names are trademarks of their owners, shown here only to report public benchmark and price data.

No favorites

How the picks are made

📊

Benchmarks, not vibes

Reasoning, coding, agentic, speed and value come from independent Artificial Analysis indices. Chat is the LMArena leaderboard — millions of blind human votes. Context and budget come from the live price catalog.

🔁

Self-updating

Nothing is hand-picked. When a new model tops a benchmark or a price changes, the cheatsheet re-ranks itself on the next daily sync. No stale "best of 2024" lists.

🎯

One axis at a time

A model can win one job and lose another. We rank each category by the single metric that matters for it, so the shortlist is honest about trade-offs.

Quality & speed from Artificial Analysis; human preference from LMArena; prices from the MyTokenTracker catalog. See the full methodology.

Citation

Use this in your work

Open data, free to cite. Pair it with the price-vs-cost breakdown and the full State of AI.

Copy a citation

Free to use and cite under CC BY 4.0. See how this is measured.

APA

Champlin Enterprises. (2026). Which AI for which job — the model cheatsheet (MyTokenTracker) [Data set]. MyTokenTracker. Retrieved July 31, 2026, from https://mytokentracker.io/which-ai

BibTeX

@misc{mytokentracker-which-ai,
  title        = {Which AI for which job — the model cheatsheet (MyTokenTracker)},
  author       = {{Champlin Enterprises}},
  year         = {2026},
  howpublished = {MyTokenTracker, \url{https://mytokentracker.io/which-ai}},
  note         = {Accessed July 31, 2026. Licensed CC BY 4.0.},
  url          = {https://mytokentracker.io/which-ai}
}

Need a fixed point in time? Every day’s data is permanently archived in the open-data repository, so you can cite a specific date by linking that day’s committed file.