Every provider we track
17 providers and 60 models, each with a full rate card and an explanation of how the lineup is tiered. Ordered below by each provider's cheapest model, because that is the number that tells you where a vendor's floor is. Verified 2026-08-16.
Entry price by provider
The cheapest model each provider sells, ranked. This is a floor, not a summary: a provider with a cheap entry tier can still be the expensive choice at the top of its lineup.
| Provider | Cheapest model | Input / 1M | Output / 1M | Models tracked |
|---|---|---|---|---|
| Alibaba | Qwen3.7 Flash | $0.03 | $0.13 | 3 |
| Groq | Llama 3.1 8B Instant (via Groq) | $0.05 | $0.08 | 2 |
| Amazon | Nova Micro | $0.035 | $0.14 | 4 |
| Cohere | Command R7B | $0.0375 | $0.15 | 3 |
| Meta | Llama 4 Scout | $0.08 | $0.30 | 2 |
| StepFun | Step 3.5 Flash | $0.10 | $0.30 | 1 |
| OpenAI | GPT-4.1 nano | $0.10 | $0.40 | 9 |
| Gemini 2.5 Flash-Lite | $0.10 | $0.40 | 7 | |
| DeepSeek | DeepSeek V4 Flash | $0.14 | $0.28 | 2 |
| Perplexity | Sonar Small Online | $0.20 | $0.20 | 4 |
| Mistral | Mistral Small 4 | $0.15 | $0.60 | 4 |
| xAI | Grok 4.1 Fast | $0.20 | $0.50 | 5 |
| Zhipu AI | GLM-5 | $0.60 | $1.92 | 3 |
| Baidu | ERNIE 5.1 | $0.56 | $2.54 | 1 |
| Moonshot AI | Kimi K2.5 | $0.60 | $3.00 | 3 |
| ByteDance | Doubao Seed 2.1 Pro | $0.85 | $4.23 | 1 |
| Anthropic | Claude Haiku 4.5 | $1.00 | $5.00 | 6 |
Provider by provider
Alibaba
Alibaba's Qwen line spans a flagship Max tier down to a Flash tier that is the cheapest model on this entire site. Rates differ sharply by region.
3 models tracked. From Qwen3.7 Flash at $0.03 in / $0.13 out, up to Qwen3.8 Max at $2.00 in / $6.00 out: a 55x internal spread. Cached-input discounts available. Batch discounts available.
Groq
Groq serves open-weight models on custom LPU hardware, competing on latency and throughput rather than model capability.
2 models tracked. From Llama 3.1 8B Instant (via Groq) at $0.05 in / $0.08 out, up to Llama 3.3 70B Versatile (via Groq) at $0.59 in / $0.79 out: a 11x internal spread. Batch discounts available.
Amazon
Amazon's Nova family is sold through Bedrock and reaches further down the price curve than most Western providers.
4 models tracked. From Nova Micro at $0.035 in / $0.14 out, up to Nova Premier 1.0 at $2.50 in / $12.50 out: a 82x internal spread.
Cohere
Cohere targets enterprise RAG and tool-use workloads, with a flagship tier and a very cheap 7B option at the bottom.
3 models tracked. From Command R7B at $0.0375 in / $0.15 out, up to Command R+ at $2.50 in / $10.00 out: a 67x internal spread.
Meta
Meta publishes Llama as open weights rather than a paid first-party API, so what you pay depends on which host you run it through, or on your own GPU bill.
2 models tracked. From Llama 4 Scout at $0.08 in / $0.30 out, up to Llama 4 Maverick at $0.20 in / $0.60 out: a 2.2x internal spread.
StepFun
StepFun is a Chinese lab shipping low-cost Flash-class models competitive with the cheapest tiers from larger providers.
1 model tracked. Step 3.5 Flash at $0.10 in / $0.30 out ($0.150 blended).
OpenAI
OpenAI sells the GPT family through a tiered lineup, from the flagship reasoning tier down to nano models built for volume, with a cached-input discount on almost every tier.
9 models tracked. From GPT-4.1 nano at $0.10 in / $0.40 out, up to GPT-5.5 Pro at $30.00 in / $180.00 out: a 386x internal spread. Cached-input discounts available.
Google's Gemini lineup runs from Pro down to Flash and Flash-Lite, with batch discounts across the board and frequent introductory pricing on new Flash releases.
7 models tracked. From Gemini 2.5 Flash-Lite at $0.10 in / $0.40 out, up to Gemini 3 Pro at $2.00 in / $12.00 out: a 26x internal spread. Cached-input discounts available. Batch discounts available.
DeepSeek
DeepSeek is among the cheapest frontier-adjacent APIs anywhere, with an unusually deep cache-hit discount and a peak/off-peak schedule.
2 models tracked. From DeepSeek V4 Flash at $0.14 in / $0.28 out, up to DeepSeek V4 Pro at $0.435 in / $0.87 out: a 3.1x internal spread. Cached-input discounts available.
Perplexity
Perplexity's Sonar models are search-grounded, which means token rates tell only half the story: per-request search fees are a real part of the bill.
4 models tracked. From Sonar Small Online at $0.20 in / $0.20 out, up to Sonar Pro at $3.00 in / $15.00 out: a 30x internal spread.
Mistral
Mistral prices aggressively for a Western provider, with a flagship tier that undercuts most competitors' mid tiers.
4 models tracked. From Mistral Small 4 at $0.15 in / $0.60 out, up to Mistral Medium 3.5 at $1.50 in / $7.50 out: a 11x internal spread.
xAI
xAI's Grok models sit in the mid-price band, with cached-input discounts and a long-context surcharge above 200K tokens.
5 models tracked. From Grok 4.1 Fast at $0.20 in / $0.50 out, up to Grok 4.5 at $2.00 in / $6.00 out: a 11x internal spread. Cached-input discounts available.
Zhipu AI
Zhipu's GLM models are among the cheapest flagship-class options tracked here, with cached input at roughly a fifth of the standard rate.
3 models tracked. From GLM-5 at $0.60 in / $1.92 out, up to GLM-5.2 at $1.40 in / $4.40 out: a 2.3x internal spread. Cached-input discounts available.
Baidu
Baidu's ERNIE models are sold through Qianfan, priced in CNY, and aggressive on cost relative to Western equivalents.
1 model tracked. ERNIE 5.1 at $0.56 in / $2.54 out ($1.06 blended).
Moonshot AI
Moonshot's Kimi models pair long context with cached-input discounts, priced well below comparable US flagship tiers.
3 models tracked. From Kimi K2.5 at $0.60 in / $3.00 out, up to Kimi K3 at $3.00 in / $15.00 out: a 5.0x internal spread. Cached-input discounts available.
ByteDance
ByteDance sells Doubao through Volcano Engine, priced in CNY and aimed primarily at the Chinese market.
1 model tracked. Doubao Seed 2.1 Pro at $0.85 in / $4.23 out ($1.70 blended).
Anthropic
Anthropic splits Claude into Fable, Opus, Sonnet, and Haiku tiers, and layers a 10% cached-input rate and a flat 50% batch discount on every one of them.
6 models tracked. From Claude Haiku 4.5 at $1.00 in / $5.00 out, up to Claude Fable 5 at $10.00 in / $50.00 out: a 10x internal spread. Cached-input discounts available. Batch discounts available.
Questions about providers
Which LLM provider is cheapest?
On entry-level tiers, Alibaba is cheapest, with Qwen3.7 Flash at $0.03 per 1M input tokens. But "cheapest provider" is the wrong question: every provider spans a wide range internally, and the gap between a vendor's own cheapest and most expensive model is usually larger than the gap between vendors at the same tier. Compare tier to tier, not brand to brand.
How many providers does this site track?
17 providers and 60 models. Each provider has a page with its full rate card, how the lineup is tiered, and which discounts apply. Every model also has its own page with worked cost examples.
Which providers offer prompt caching discounts?
27 of the 60 models tracked publish a discounted cached-input rate. It is the single biggest cost lever for any workload that resends a stable prefix, and the discount is typically 80-90% off the standard input rate. The prompt caching savings calculator sizes it against your hit rate.
Which providers offer batch discounts?
18 models tracked here publish a batch-API discount, usually a flat 50% off both input and output for asynchronous work. If your workload can tolerate delay, that is a straight halving of the bill with no change to the model.
Are Chinese providers really cheaper?
Consistently, yes, and by a wide margin at every tier. DeepSeek, Alibaba, Zhipu, Moonshot, ByteDance and Baidu all sit well below the US price band. Price is only one input to the decision though: data residency, latency from your region, and API reliability all matter, and none of them show up in a rate card.