COMPARISON · VERIFIED 11 JUL 2026
Which is cheapest: Claude, GPT, or Gemini?
Claude is cheapest for most use cases, with Haiku 3.5 costing $0.80 per million input tokens compared to GPT-4o mini at $0.15 and Gemini 1.5 Flash at $0.075 per million tokens—though actual cost depends on your prompt size, output length, and caching strategy. For flagship models, pricing is nearly identical across providers at $3-15 per million tokens.
The honest answer is "it depends which tier you're comparing" — and most head-to-head posts pick one model from each provider without saying why that one. Here's the comparison broken out by tier, so you can compare like for like instead of a cherry-picked flagship-vs-budget matchup.
Budget tier
Cheapest way to run high volume, low-complexity tasks
At the budget tier, OpenAI and Google are priced identically and both undercut Anthropic's Haiku by a wide margin. If your workload is simple classification, extraction, or high-volume low-stakes generation, GPT-4o mini or Gemini 2.5 Flash are the cheaper starting point — the quality gap on straightforward tasks is usually smaller than the price gap.
Mid tier
The default production choice for most teams
This is where most production workloads actually live, and the gap narrows. Claude Sonnet 5 is at introductory pricing through 31 August 2026, after which it moves to $3/$15 — worth factoring in if you're budgeting past that date. Gemini 3 Flash remains the clear value pick if your task doesn't need the extra reasoning quality Sonnet and GPT-4o are priced for.
Flagship tier
Maximum capability, priced accordingly
Gemini 3.1 Pro undercuts both Opus 4.8 and GPT-5.5 on paper while offering the largest context window of the three (2M tokens). Opus and GPT-5.5 sit close together on input price, with GPT-5.5 slightly more expensive on output. At this tier, the decision is rarely about the token price alone — it's about which model's reasoning style fits your specific task, since the cost difference on most real workloads is smaller than the headline per-token gap suggests.
The number that actually matters: your workload, not the rate card
A flagship model that gets a task right in one pass can be cheaper in practice than a budget model that needs three retries and a longer prompt to get there. Use the calculator with your own rough input/output split before assuming the cheapest per-token rate is the cheapest actual bill.
* Claude Sonnet 5 pricing shown is introductory through 31 Aug 2026; standard $3.00/$15.00 pricing applies from 1 Sep 2026. All prices verified against official provider documentation, 11 July 2026 — see full methodology.