Gemini Flashcheapest
Google · $0.1/M in · $0.4/M out
$0.00036
The cheapest serious option, with a very large context window.
55,555 calls per budget · 1.5× cheaper than GPT-4o mini
Typical call size
Medium means a page of context: 800 tokens in, 700 out.
Google · $0.1/M in · $0.4/M out
$0.00036
The cheapest serious option, with a very large context window.
55,555 calls per budget · 1.5× cheaper than GPT-4o mini
Meta / hosts · $0.3/M in · $0.4/M out
$0.00052
Open weights via a host — cheap output, quality varies by provider.
38,461 calls per budget · 1.0× cheaper than GPT-4o mini
OpenAI · $0.15/M in · $0.6/M out
$0.00054
The default workhorse: fast, cheap, good enough for most everyday prompts.
37,037 calls per budget · the ChatGPT baseline
OpenAI · $0.4/M in · $1.6/M out
$0.00144
A step up in instruction-following for a few times the price of 4o mini.
13,888 calls per budget · 2.7× GPT-4o mini
Anthropic · $0.8/M in · $4/M out
$0.00344
Quick and tidy writer; strong at following formatting rules.
5,813 calls per budget · 6.4× GPT-4o mini
OpenAI · $2/M in · $8/M out
$0.00720
Long-context, high-accuracy work: contracts, code, careful reasoning.
2,777 calls per budget · 13.3× GPT-4o mini
Google · $1.25/M in · $10/M out
$0.00800
Google's reasoning tier, priced close to GPT-4o on output.
2,500 calls per budget · 14.8× GPT-4o mini
OpenAI · $2.5/M in · $10/M out
$0.00900
The flagship multimodal model — text, images and audio in one place.
2,222 calls per budget · 16.7× GPT-4o mini
Anthropic · $3/M in · $15/M out
$0.0129
A favourite for long-form drafting and code review; the priciest here.
1,550 calls per budget · 23.9× GPT-4o mini
USD list prices per 1M tokens. Published rates change often — use these as a planning estimate, not a quote.
Most models charge three to five times more for the answer than the prompt. Capping reply length with max_tokens saves more money than trimming your prompt.
A medium call on GPT-4o mini lands at $0.00054. Gemini Flash undercuts it slightly; Claude Sonnet costs many times more for the same words.
Route the easy 90% — classification, tidy-ups, summaries — to a mini model, and reserve a flagship for the calls where being wrong is expensive.
In a chat app you resend the whole conversation every turn, so cost grows with the square of the thread length. Summarise old turns and the curve flattens.