Among the current models we recommend, Mistral Small 3.1 has the lowest list price at $0.10/$0.30 per million tokens. It is not the best cheap AI for most people, though. Claude Haiku 5.5 and GPT-6 Luna cost $0.10/$0.50 — only slightly more on output — and are current-generation models from Anthropic and OpenAI, so they do far more useful work per dollar.
The safest this comparison default, the cheaper option worth trying first, and the specialist pick — before you read the detail below.
GPT-6 Luna is the strongest answer here for this comparison — pick it when quality of output matters more than the $0.10/1M/1M input you pay for it.
Mistral Small 3.1 handles the same job for about 33% less per token. Start here and only move up if the output is not good enough.
Claude Haiku 5.5 is the fastest of these for this comparison — worth it when latency is what the reader notices, not the last few points of reasoning depth.
Mistral Small 3.1 is the cheapest by list price among the models we recommend, at $0.10/$0.30.
Claude Haiku 5.5 and GPT-6 Luna are the best cheap defaults at $0.10/$0.50.
The lowest price is not the same thing as the best value — retries and rework cost money too.
Use the absolute cheapest model for internal, review-heavy, low-stakes tasks.
Use Claude Haiku 5.5 or GPT-6 Luna when the work still needs to be broadly useful.
Use open-weight models such as DeepSeek V4.1 Flash or Qwen 3.8 Flash when you may want to self-host later.
Switch the scoring lens to see whether the this comparison answer changes when cost, speed, or long-document depth leads the decision.
DeepSeek / Budget / Oct 10, 2026
DeepSeek's open, cheap new default — V4-Pro now routes to it.
Ranks models by the broadest mix of coding, writing, research, and long-context usefulness.
You need long single outputs (32K cap) or independently verified scores before you commit.
Where the this comparison recommendation shifts once you weigh price or latency differently.
OpenAI's budget tier at half the old price.
Anthropic's budget model, now with adjustable reasoning, at $0.10/$0.50.
Ultra-cheap multimodal model for massive-volume, low-complexity pipelines.
SWE-bench Pro 62.5 at sixteen cents per million input.
DeepSeek's open, cheap new default — V4-Pro now routes to it.
List prices and published scores — the numbers this page's pick is built from.
| Model | Input | Output | Est. month | Context | Speed | Coding | Writing | Research |
|---|---|---|---|---|---|---|---|---|
| GPT-6 LunaOpenAI | $0.10/1M | $0.50/1M | $2.00 | 1.1M tokens | Very fast | 84 | 82 | 80 |
| Claude Haiku 5.5Anthropic | $0.10/1M | $0.50/1M | $2.00 | 1M tokens | Very fast | 86 | 84 | 82 |
| Mistral Small 3.1Mistral | $0.10/1M | $0.30/1M | $1.60 | 128k tokens | Very fast | 55 | 66 | 52 |
| Qwen 3.8 FlashAlibaba | $0.15/1M | $0.47/1M | $2.44 | 991k tokens | Very fast | 84 | 76 | 78 |
Scores out of 100 — how we evaluate models. “Est. month” is 10M in / 2M out at list price: a ceiling, no discounts.
Why each one is on the shortlist for this comparison, what it is genuinely good at, and where we would steer you away from it.
The default answer for this comparison — 84/100 on the coding axis, and the model we would start with unless the price below rules it out.
OpenAI's September 22, 2026 low-cost reasoning model, replacing GPT-5.6 Luna at $0.10/$0.50 per 1M — half the predecessor's price, with the same 1.05M context.
You are choosing a budget model from scratch — Claude Haiku 5.5 costs the same and scores higher on Artificial Analysis's index.
A cheaper Luna rather than a smarter one. At $0.10/$0.50 it matches Claude Haiku 5.5 on price, but independent scoring puts Haiku 5.5 slightly ahead; choose Luna when you are already on the OpenAI stack.
Full pricing, benchmark table and release notes on the GPT-6 Luna page.
The fastest model in this shortlist for this comparison. Pick it when turnaround is what your readers or users notice.
Anthropic's October 7, 2026 low-cost model and the first Haiku with adjustable reasoning effort — thinking can be switched off for simple requests or raised to max, at $0.10/$0.50 per 1M under 100K input tokens.
Your prompts routinely exceed 100K tokens, where the price quintuples, or the task needs sustained reasoning that Sonnet 5.5 handles far better.
The cheapest way to put an Anthropic model in a high-volume pipeline. Keep requests under 100K tokens to stay on the low rate, and set effort per task — at low effort it is a fast classifier, at max it handles real tool-use work.
Full pricing, benchmark table and release notes on the Claude Haiku 5.5 page.
The cost-conscious pick for this comparison, about 33% less per token than GPT-6 Luna than the top choice while holding 55/100 on coding.
Ultra-cheap multimodal model for massive-volume, low-complexity pipelines. Full Mistral Small 3.1 review →
Also worth a look for this comparison, at 84/100 on the coding axis.
SWE-bench Pro 62.5 at sixteen cents per million input. Full Qwen 3.8 Flash review →
UseRightAI recommendations are based on practical decision factors people actually feel in day-to-day use.
Newsletter
We email when the this comparison pick changes, when one of these models moves on price, or when something new displaces the current leader.
No spam. Useful updates only. Affiliate disclosures always clearly labeled.
Among the current models we recommend, Mistral Small 3.1 at $0.10/$0.30 per million tokens. Claude Haiku 5.5 and GPT-6 Luna cost $0.10/$0.50, and GPT-6 Luna's Flex tier halves that for work that can wait.
Claude Haiku 5.5 and GPT-6 Luna. Haiku 5.5 scores 43 on Artificial Analysis's Intelligence Index and Luna 38 — both far above older budget models. Haiku 5.5's price rises fivefold above 100K input tokens, so keep prompts short.
DeepSeek V4.1 Flash, at about $0.30/$1.20 through hosted APIs, with MIT-licensed weights; DeepSeek says it beats its own V4-Pro on coding. Qwen 3.8 Flash is cheaper still at $0.15/$0.47.
No. If poor output causes rework or mistakes, a slightly more expensive model can be the cheaper operational decision.
Because providers quote input and output separately, and output almost always costs several times more. A model advertised at a very low input price can still produce a larger bill than a pricier rival if it is verbose, because your spend tracks the length of the responses rather than the length of the prompts. Compare both numbers against a realistic workload before deciding — a moderate month of 10M input and 2M output tokens is a useful common basis, and the figure it produces is the one worth ranking on.
Retries and rework, mostly. A weaker model that needs two attempts at the same task has doubled its real price, and it has also spent your time. Reasoning models add a subtler version: the hidden thinking tokens are billed as output on most providers, so a cheap-looking reasoning model can cost more per completed task than an expensive non-reasoning one. Judge cost per finished piece of work, not cost per million tokens.
A lot, because most tokens in a real workload are not hard problems. Classification, extraction, routine drafting, and intermediate agent steps make up the bulk of the volume and rarely need a frontier model. Routing those to a budget model and reserving a premium one for genuinely difficult calls typically cuts spend substantially without a quality drop anyone notices. It is more work to build than picking one model, and it is usually the single largest saving available.