Which model for your job
The cheapest model that is good enough for your job, and the next ones in line. Say what the job reads and writes, how often it runs, how hard it is and whether it can wait. Each model's independent capability score sets the bar; its price sets the order.
Your job
Start from the job nearest yours, then change its size and runs a day. The shortlist keeps each maker's newest models whose independent score clears the bar for that kind of work and that can take the job, cheapest first.
A short question and a few paragraphs back.
Chat, writing, summaries and support replies. Models that score at least as high as GPT-5.
about 230 words
about 380 words
Instructions or documents the model has read before cost far less the second time.
Try first
Claude Haiku 5.5
84 cents
Anthropic · a month of this job · 0.028 cents a run
- The cheapest of 14 models that score at least as high as GPT-5 that can take this job.
- Scores 159 on the Epoch Capabilities Index (157 to 162), where GPT-5 is 150.
- Its largest prompt and answer, 800 tokens, fit its 1M window.
- GPT-6 Luna costs the same for this job.
Next in line
- GPT-6 LunaOpenAI · score 156 · the same bill84 cents
- DeepSeek V4.1 FlashDeepSeek · score 156 · 2.5 times the bill$2.07
- Gemini 3.8 FlashGoogle · score 157 · 7.5 times the bill$6.30
- DeepSeek V4 ProDeepSeek · score 155 · 8.5 times the bill$7.13
- Qwen3.8-MaxAlibaba · score 157 · 13 times the bill$10.80
And 8 more that cost more. Each one opens its comparison with Claude Haiku 5.5.
The cheapest model for 8 common jobs
Each job at its usual runs a day for a month, among the newest models whose score clears the bar for its kind of work, counted in tokens. The score is the Epoch Capabilities Index, where GPT-5 is 150. The last column is the cheapest when the answers can wait a day.
| Job | Kind of work | Try first | Then | If it can wait |
|---|---|---|---|---|
| A chat question100 runs a day | Everyday | Claude Haiku 5.584 cents a month, score 159 | GPT-6 Luna84 cents a month, score 156 | Claude Haiku 5.542 cents a month, batch |
| A hard question, with thinking100 runs a day | Hard | GPT-6.1 Sol$43.80 a month, score 166 | Gemini 4 Argon$43.80 a month, score 165 | GPT-6.1 Sol$21.90 a month, batch |
| A long report, summarised20 runs a day | Everyday | Claude Haiku 5.5$4.20 a month, score 159 | GPT-6 Luna$4.20 a month, score 156 | Claude Haiku 5.5$2.10 a month, batch |
| A 40-turn agent task10 runs a day | Hard | GPT-6.1 Sol$364 a month, score 166 | Gemini 4 Argon$364 a month, score 165 | GPT-6.1 Sol$182 a month, batch |
| A support bot's reply2,000 runs a day | Everyday | Claude Haiku 5.5$22.20 a month, score 159 | GPT-6 Luna$22.20 a month, score 156 | GPT-6 Luna$11.10 a month, batch |
| Sort a message into a category50,000 runs a day | Simple | Claude Haiku 5.5$41.25 a month, score 159 | GPT-6 Luna$41.25 a month, score 156 | Claude Haiku 5.5$20.63 a month, batch |
| Read an invoice into fields1,000 runs a day | Simple | Claude Haiku 5.5$8.25 a month, score 159 | GPT-6 Luna$8.25 a month, score 156 | Claude Haiku 5.5$4.12 a month, batch |
| Review a pull request50 runs a day | Hard | GPT-6.1 Sol$143 a month, score 166 | Gemini 4 Argon$143 a month, score 165 | GPT-6.1 Sol$71.25 a month, batch |
How the shortlist is made, and what it cannot tell you
First, the bar. Each model's quality comes from the Epoch Capabilities Index by Epoch AI, an independent research group: one score that combines over 50 benchmarks, scaled so that GPT-5 is 150 and Claude 3.5 Sonnet 130. For hard work a model must score as high as the best of the newest models; for everyday work, at least as high as GPT-5; for simple work, any model will do. Every score has a range of doubt, and a model clears the bar when the top of its range reaches it, so 2 models that cannot be told apart are treated alike.
Then, whether a model can take the job at all. The prompt and the answer must fit in its context window, and a single answer cannot be longer than the most the model writes at once. A model that fails either test is set apart and the reason is named.
Last, the price. The tool prices 1 run with each maker's own prices, the cache price for text read again, the long-prompt price where a prompt passes the maker's limit, and the batch price when the work can wait a day, then multiplies by the runs a day for a month of 30 days. Counted in words, the bill also allows for makers that cut the same text into more tokens. Cheaper models that miss the bar are listed too, so you can see what the bar costs.
One score cannot know your work. A model can rank high overall and still be weak at your language, your format or your field, and a model with no independent score yet is untested here, not bad. Take the first 2 or 3 on the shortlist, run the same 10 real tasks on each, and keep the cheapest one whose answers you would accept.
Questions
- Which AI model should I use?
- The cheapest one that is good enough for the job. Hard work such as code, agents and long reasoning needs a model whose independent score is as high as the best; chat, writing and summaries need one at least as good as GPT-5; sorting and reading fields out of documents can use any model. Among those that fit your prompt, try the cheapest first. This tool does that sum for your own job.
- What is the cheapest model for a support bot?
- At 2,000 replies a day, among models good enough for everyday work, Claude Haiku 5.5 from Anthropic and GPT-6 Luna cost the same and least, about $22.20 a month. The dearest of them, GPT-6 Astra, would cost $2,220, 100 times as much.
- What is the cheapest model for an AI agent?
- For 10 agent tasks a day of 40 turns each, with 90 percent of the input read from cache, GPT-6.1 Sol from OpenAI and Gemini 4 Argon cost the same and least of the models that score as high as the best, about $364 a month. GPT-6 Astra would cost $2,196, 6 times as much.
- What is the cheapest model for sorting messages into categories?
- At 50,000 messages a day, each a short message in and one word out, Claude Haiku 5.5 from Anthropic and GPT-6 Luna cost the same and least of any newest model, about $41.25 a month. If the sorting can wait a day, Claude Haiku 5.5 on a batch costs $20.63.
- Is the cheapest model good enough for my job?
- It is good enough on average for the kind of work you chose: the shortlist only keeps models whose independent score clears the bar for it. An average is not your job, so run 10 real tasks on the first 2 or 3 models and keep the cheapest one whose answers you would accept. The prices often differ several times over, so the test pays for itself.
- Which AI model is the smartest right now?
- On the Epoch Capabilities Index, Claude Opus 5.5 from Anthropic scores highest of the newest models we list, at 167. GPT-6 Astra, GPT-6.1 Sol, Claude Fable 5.1, Gemini 4 Argon and Claude Sonnet 5.5 score within its range of doubt, so they cannot be told apart from it. The cheapest of that group for a chat question is GPT-6.1 Sol.
- Does it cost less if the answers can wait?
- On some models, by about half. OpenAI and Anthropic list a batch price for work that can wait up to a day, about half the usual price. Where a maker lists none, the tool charges the usual price either way, so the order of the shortlist can change when you say the work can wait.
Put this calculator on your own site
Copy the code into any web page: a blog post, a course page or your team's wiki. The calculator runs in a frame and shows the same figures as this page, so when the prices change here they change on your page too. It is free and needs no account, and nothing a reader types is sent anywhere.
The code opens the calculator with the settings you have chosen above.
The short script at the end lets the frame grow and shrink with the calculator. Where a page cannot take scripts, choose a fixed height and the frame scrolls.
<iframe src="https://tecyfy.com/embed/which-model" title="Which model for my job" width="100%" height="2500" style="border:0;display:block" loading="lazy" allow="clipboard-write"></iframe>
<p><a href="https://tecyfy.com/prices/which-model">Which model for my job</a> by Tecyfy</p>
<script>addEventListener("message",function(e){if(e.origin!=="https://tecyfy.com"||!e.data||e.data.type!=="tecyfy:height")return;var h=Number(e.data.height);if(!(h>0))return;document.querySelectorAll("iframe").forEach(function(f){if(f.contentWindow===e.source)f.style.height=Math.min(h,20000)+"px"})})</script>