Updated October 18, 2026 · DeployCue Editorial Team
Top LLM APIs compared
We earn a commission when you sign up through links on this site. Partner status can affect where a brand appears in our rankings. How we rank
| # | Provider | Score | Features | Key facts | Actions |
|---|---|---|---|---|---|
| 1 |  BytePlusTop pick Model: pay-per-use | 8.7/10 |
|
| ReviewCompareVisit |
| 2 |  Zhipu AI (Z.ai) | 9.4/10 |
| ReviewCompareVisit | |
| 3 |  WaveSpeedAI Model: pay-per-use | 8.8/10 |
|
| ReviewCompareVisit |
| 4 |  Azure AI Foundry Pay as you go or try Azure free for up to 30 days | 9.6/10 |
| ReviewCompareSee alternatives | |
| 5 |  Salad Model: Flat monthly subscription or pay per token | 9.6/10 |
|
| ReviewCompareSee alternatives |
| 6 | Zhipu AI Cache storage free for a limited time: CNY 0 per 1M tokens per hour (cache storage currently free of charge) | 9.6/10 |
| ReviewCompareSee alternatives | |
| 7 |  AI/ML API Model: Pay-as-you-go per-model pricing per 1M tokens on a prepaid balance; minimum top-up $20 | 9.5/10 |
| ReviewCompareSee alternatives | |
| 8 |  Amazon Web Services AWS Business Support free trial: AWS Business Support at no cost for up to 60 days | 9.5/10 |
|
| ReviewCompareSee alternatives |
| 9 |  IBM watsonx Free trial of watsonx.ai (free Toolbox playground): Up to 300,000 foundation model tokens, 20 CUH of machine learning tools and 100 text extraction documents per month at no cost | 9.5/10 |
|
| ReviewCompareSee alternatives |
| 10 |  Parasail | 9.5/10 |
|
| ReviewCompareSee alternatives |
| 11 |  Amazon Web Services AWS Free Tier credits for new customers: $100 immediately, up to $100 more, up to $200 total | 9.4/10 |
|
| ReviewCompareSee alternatives |
| 11 |  Cloud.ru (Sber Cloud) Up to 20,000 bonus rubles on registration: 4,000 RUB for individuals, 20,000 RUB for legal entities and sole proprietors | 9.4/10 |
|
| ReviewCompareSee alternatives |
| 12 |  DeepInfra 50% off zai-org/GLM-5.3-Flash | 9.4/10 |
|
| ReviewCompareSee alternatives |
| 13 |  Groq | 9.4/10 |
| ReviewCompareSee alternatives |
About Top LLM APIs compared
LLM API providers are compared here on the three things that decide a monthly bill: what you pay when you start, what you pay per token once traffic builds, and what the plan covers. This ranking spans 127 brands, from a gateway that fronts other model providers to a hyperscale cloud selling its own catalogue. The main trade-off is a low entry cost against a per-token floor that can climb steeply. Input price floors run €0 to €1.32, and the median input floor is €0.08. Output price floors run €0 to €5.26, and the median output floor is €0.23. The median entry price is €3.07, with the range running €1.93 to €6.95. Two providers can both advertise a small starting credit and settle very different invoices. Every entry serves an OpenAI-compatible endpoint, so the choice narrows to catalogue, limits and trust, with a median context window ceiling of 1,000,000 tokens.
How we picked
Every entry is scored on published prices, offer terms and trust signals, and the weights are fixed. Cost carries 40%, the offer weighting is 40% and trust accounts for 20%, so price at entry and at renewal does most of the work while certifications and the contracting entity carry real but smaller weight. Figures come from each provider's own documentation, pricing pages, terms and security pages, plus company records for founding year, headquarters and legal entity, and from site-wide statistics published here where a field is not published per brand. Where a provider states a free credit, a context window limit or a support channel, that statement is used as given and keeps its qualifier. Brand names and legal entities are written as the providers record them, and comparisons are drawn only from those records. Partners that serve your country are listed first in this ranking, then the rest by score. The pool behind the ranking holds 127 brands, drawn from 127 entries in this category.
How we rankOur top picks
4. Azure AI FoundryAzure AI Foundry sits in this ranking on catalogue breadth and enterprise paperwork rather than on the lowest price. Its certification list, ISO 27001, ISO 27018 and SOC 1, is the longest here, and the model list, gpt 6 astra, gpt 6 luna and gpt 6 sol, spans instruction, reasoning, embedding, image, audio and video families. New accounts receive USD 200 in Azure credits, and the largest context window reaches 1,050,000 tokens. Support runs on help centre and community, a help centre and community, so UCloud below reads as the lighter, more reachable alternative for teams that weigh regions before catalogue depth.
- Models offered: GPT-6-astra, GPT-6-luna, GPT-6-sol, and 39 more
- Largest context window: 1,050,000
- Free credits: USD 200 in Azure credits
Pros
- The catalog in gpt 6 astra, gpt 6 luna and gpt 6 sol is far wider than Salad's qwen3.6 35b a3b, qwen3.6 27b and qwen3.5 9b.
- A 1,050,000 token window takes very long documents in one call.
- USD 200 in Azure credits gives room to run real prompts before spending.
Cons
- Support runs on help centre and community only, so no phone or chat line is listed.
- The catalog in gpt 6 astra, gpt 6 luna and gpt 6 sol is OpenAI-led, while Supascale AI covers more provider families in DeepSeek V4 Pro, GPT 5.6 Sol and GPT 5.6 Terra.
- On the largest context window this brand ranks 5 of 72 brands.
5. SaladSalad earns its place in this ranking by pairing an OpenAI-compatible gateway with a stated compliance posture, and it is the older operation of the three, founded in 2018 and headquartered in US under Salad Technologies, Inc.. Its certification is SOC 2 Type I, one attestation to read closely rather than a broad list. Support runs on email, help centre and community, adding a community channel to email and a help centre. Azure AI Foundry below spreads across a far wider set of certifications, so the comparison is one named attestation against many.
- GPU models: NVIDIA RTX 3060, NVIDIA RTX 3070, NVIDIA RTX 3070 Ti, and 6 more
- GPU hour from: $0.015
- Billing granularity: Per second
Pros
- Billing is per second against an hourly model, so short runs and interrupted jobs pay only for the time they hold a card.
- Workloads sit in from 190 locations, well above the 6 median in the GPU ranking.
- The catalogue in NVIDIA RTX 3060, NVIDIA RTX 3070 and NVIDIA RTX 3070 Ti spans several RTX generations, the RTX PRO workstation class and selected AMD classes.
- The AI gateway is OpenAI-compatible, so clients already written against that API can be repointed at Salad.
- The operator is Salad Technologies, Inc., a US company founded in 2018, with SOC 2 Type I listed for security.
Cons
- One GPU per node: None - single-GPU nodes, no multi-GPU NVLink interconnect, so any plan that needs several accelerators inside a single machine has to look elsewhere.
- Bring your own key is not offered, and the gateway is billed as Flat monthly subscription or pay per token, so you cannot pay through your own provider account.
- Support runs on email, help centre and community only, with no live chat or phone line listed for urgent issues.
- A free tier is not offered, and only 27% of the GPU ranking sits in the same paid-only position.
What to look for
- Entry price and what it buys
- Entry price is the first number a reader meets, and it varies widely: the median is €3.07, the lower quartile €2.63 and the upper quartile €4.18. Check whether that figure is free credits, a daily request allowance or a prepaid balance, because the three behave differently once a month turns over.
- Input and output price floors
- The recurring cost is the per-token floor. Output price runs €0 to €5.26, and output usually decides the invoice for a chat product. The median input floor is €0.08, so weight output more heavily when comparing two providers.
- Context window and long documents
- Context window decides how much you can send in one request. The median maximum here is 1,000,000 tokens, with 256,000 at the low end and a ceiling of 10,000,000. A shorter window means building retrieval and chunking before the model sees your data.
- What the catalogue covers
- Model breadth tells you whether one endpoint can carry text, image, audio and video work. Some entries serve a single instruction-model line, while others run the full spread behind one OpenAI-compatible API. A wider catalogue stretches from very cheap small models to premium tiers, so check the specific models you plan to call.
- Trust signals behind the endpoint
- Certifications and the contracting entity tell you who stands behind the data. 127 entries appear in this category and trust carries 20%, below cost and offer, but a named legal entity and a recognised certification are worth reading before production traffic starts.
Frequently asked questions
What does an OpenAI-compatible API change for a team?+
An OpenAI-compatible API lets you keep your existing client library and change the base URL and key, so current code runs against another provider with few edits. The ranking records this as a plain fact for each brand, and the entries that carry it sit more easily beside OpenAI-based codebases. A provider that keeps its own request format means rewriting the client layer before the first response arrives.
Should I weigh the entry price or the per-token price?+
Entry price is the number you meet first, often free credits or a small prepaid balance, and it says little about what a busy month costs. Per-token input and output prices recur, and they vary widely across this ranking. Read the entry figure for the trial, then read the input and output price floors for the renewal, because that is where spend accumulates once traffic is real.
How much does the context window matter for a new project?+
Context window is the longest prompt a provider accepts in one request, and it decides how much you can send without summarising or chunking. In this ranking the typical maximum sits at 1,000,000 tokens, with 10,000,000 at the far end. A short window means extra engineering around retrieval and memory before the model ever sees your text.
Which support channels should I expect from a provider?+
Support reaches you by different routes, and a team with a live product needs more than a forum thread. The most common route here is a help centre at 84.5%, with community at 81% and email at 76.7%. Only a small share of entries add chat or phone, so check the channels listed for your pick before you commit.
What kinds of free access appear across the ranking?+
Free access differs sharply across the ranking. One entry hands out a daily allowance of requests, another gives a fixed token bundle, another offers a credit after you contact support with a card linked, and another asks for a minimum prepaid purchase before any call goes through. A small token count suits a first test, while a daily allowance suits a team running prototypes all week.
Does a bigger model catalogue mean better value?+
Model lists tell you whether a provider covers your workload, from open-weight instruction models to image, video and speech generation. Several entries here run text, image, audio and video models behind one OpenAI-compatible endpoint. Check breadth against cost, because a wide catalogue also stretches from very cheap small models to premium tiers, and the rate you actually pay depends on the model you call.