The cheapest LLM APIs, November 2026
Updated
As of 2026-11-15, the cheapest quality-filtered LLM API is Mistral-Nemo-Instruct-2407 at DeepInfra: USD 0.029 / USD 0.03 per 1M input / output tokens. Cheapest premium-priced model (maker output price USD 10+): GPT-5 (2025-08-07) at USD 1.25 / USD 10 (OpenAI). From a maker's own API: Qwen3.7 Flash at USD 0.03 / USD 0.13. Prices from each provider's own page.
Prices read from 41 providers' own pricing pages, newest check 2026-11-15; each model page links the source of every row. BroadRouter is run by BroadNet; BroadNet's prices are shown at the end on the same blend but not ranked.
Ranking: a blend of 3 input tokens to 1 output token, in USD per 1M tokens. That weighting fits chat and retrieval; an output-heavy job such as long reasoning favours a low output price instead, so check both columns.
Rules: Only models with a standard price from 2 or more providers are ranked. Each is ranked on its cheapest standard, paid price. With 3 or more such prices, one under a third of the model's median price on input or output counts as an outlier unless another provider charges the same (within 5%): it stays on the model page but does not set the rank here. Batch, flex or spot, priority, fast and ultrafast tiers are shown on model pages but never ranked. Free rows are not ranked. Prices a provider calls promotional or discounted rank as published and are tagged "promo". DeepSeek rows use its peak-hour price. On this check the filter sets aside 230 single-provider models and a lower outlier price on 8 models; each outlier is named in its row so you can check it.
Jump to: premium-priced models · mid-priced models · small models · makers' own APIs · where BroadNet stands. The same data as open CSV and JSON: the LLM API price index and the full provider price database, with a source and quote on every price.
What is the cheapest LLM API overall?
The 10 lowest blended prices among 198 models that pass the rules.
| # | Model | Cheapest provider | Input | Output | Blended | Context |
|---|---|---|---|---|---|---|
| 1 | Mistral-Nemo-Instruct-2407 Mistral AI | DeepInfra | USD 0.029 | USD 0.03 | USD 0.02925 | 131,072 |
| 2 | GPT OSS 20B OpenAI | AkashML promo | USD 0.018 | USD 0.09 | USD 0.036 | 131,072 |
| 3 | Gemma 3 4B Google | Amazon Bedrock | USD 0.04 | USD 0.08 | USD 0.05 | not stated |
| 4 | Llama 3 8B Stheno V3.2 Sao10K | Novita AI | USD 0.05 | USD 0.05 | USD 0.05 | 8,192 |
| 5 | Qwen 2.5 7B Qwen | SiliconFlow | USD 0.05 | USD 0.05 | USD 0.05 | 32,768 |
| 6 | Mistral-Small-24B-Instruct-2501 Mistral AI | DeepInfra | USD 0.05 | USD 0.08 | USD 0.0575 | 32,768 |
| 7 | Qwen3-8B Qwen | SiliconFlow | USD 0.06 | USD 0.06 | USD 0.06 | 131,072 |
| 8 | Llama 3.2 1B Instruct Meta | Cloudflare Workers AI | USD 0.027 | USD 0.201 | USD 0.0705 | not stated |
| 9 | Gemma 3 12B Google | DeepInfra | USD 0.05 | USD 0.15 | USD 0.075 | 131,072 |
| 10 | IBM Granite 4.2 8B IBM | CoreWeave Forge | USD 0.06 | USD 0.15 | USD 0.0825 | 131,072 |
Report a wrong price or number (email to info@broadnet.ai, with this page's link filled in).
Cheapest premium-priced models
Rule: The maker's own output price is USD 10 or more per 1M tokens (or, with no maker price, the median standard output price is). This band is defined by price alone, not by capability or release date, so it can include an older model its maker still sells at flagship prices; check the model's age before you treat it as current. The cheapest host shown can be lower than the maker. Models whose maker marks them deprecated or retiring in their price row are left out of these three bands (4 on this check); they stay in the overall ranking. 34 models qualify; the 10 cheapest are shown.
| # | Model | Cheapest provider | Input | Output | Blended | Maker's own price |
|---|---|---|---|---|---|---|
| 1 | GPT-5 (2025-08-07) | OpenAI · 3 providers | USD 1.25 | USD 10 | USD 3.438 | USD 1.25 / USD 10 (OpenAI) |
| 2 | GPT-5.1 | OpenAI · 2 providers | USD 1.25 | USD 10 | USD 3.438 | USD 1.25 / USD 10 (OpenAI) |
| 3 | Kimi K3 | Chutes · 18 providers | USD 1.85 | USD 9.50 | USD 3.763 | USD 3 / USD 15 (Moonshot AI) |
| 4 | Kimi K3.1 Pro | DeepInfra promo · 13 providers | USD 1.90 | USD 9.50 | USD 3.80 | USD 2 / USD 10 (Moonshot AI) |
| 5 | Claude Sonnet 5 | Anthropic · 5 providers | USD 2 | USD 10 | USD 4 | USD 2 / USD 10 (Anthropic) |
| 6 | Claude Sonnet 5.5 | Anthropic · 6 providers | USD 2 | USD 10 | USD 4 | USD 2 / USD 10 (Anthropic) |
| 7 | Gemini 4 Argon | Google Gemini API promo · 4 providers | USD 2 | USD 10 | USD 4 | USD 2 / USD 10 (Google Gemini API) |
| 8 | Gemini 4 Carbon | Google Gemini API promo · 4 providers | USD 2 | USD 10 | USD 4 | USD 2 / USD 10 (Google Gemini API) |
| 9 | GPT-6 Sol | OpenAI · 6 providers | USD 2 | USD 10 | USD 4 | USD 2 / USD 10 (OpenAI) |
| 10 | GPT-6.1 Sol | OpenAI · 6 providers | USD 2 | USD 10 | USD 4 | USD 2 / USD 10 (OpenAI) |
Cheapest mid-priced models
Rule: Neither premium-priced nor small: most large open-weight models and mid-tier closed models. Models whose maker marks them deprecated or retiring in their price row are left out of these three bands (4 on this check); they stay in the overall ranking. 100 models qualify; the 10 cheapest are shown.
| # | Model | Cheapest provider | Input | Output | Blended | Maker's own price |
|---|---|---|---|---|---|---|
| 1 | Mistral-Nemo-Instruct-2407 | DeepInfra · 2 providers | USD 0.029 | USD 0.03 | USD 0.02925 | — |
| 2 | DeepSeek V4 Flash 0731 | DeepInfra · 15 providers | USD 0.06 | USD 0.18 | USD 0.09 | — |
| 3 | Ling 3.0 Flash | DeepInfra · 2 providers | USD 0.06 | USD 0.18 | USD 0.09 | — |
| 4 | Ling 3.0 Flash VL | DeepInfra · 2 providers | USD 0.06 | USD 0.18 | USD 0.09 | — |
| 5 | OpenAI GPT OSS 120B | Crusoe Cloud · 21 providers · outlier left out: CoreWeave Forge USD 0.03 / USD 0.17; DeepInfra USD 0.037 / USD 0.17; AkashML USD 0.037 / USD 0.187 | USD 0.05 | USD 0.25 | USD 0.10 | — |
| 6 | DeepSeek V4 Flash | DeepInfra · 8 providers | USD 0.09 | USD 0.18 | USD 0.1125 | — |
| 7 | GLM 5.3 Flash | GMI Cloud promo · 20 providers | USD 0.075 | USD 0.25 | USD 0.1187 | USD 0.15 / USD 0.50 (Z.ai) |
| 8 | MiMo V2.6 Flash | GMI Cloud promo · 5 providers | USD 0.112 | USD 0.224 | USD 0.14 | — |
| 9 | Step-3.5-Flash | Featherless · 2 providers | USD 0.10 | USD 0.30 | USD 0.15 | — |
| 10 | Qwen3 235B A22B Instruct 2507 FP8 | GMI Cloud promo · 5 providers | USD 0.0875 | USD 0.35 | USD 0.1531 | — |
Cheapest small models
Rule: 35B total parameters or fewer by the model's name (or, where the name has no size, the maker's model card), or a name that marks a small model (mini, nano, lite, small, haiku). Models whose maker marks them deprecated or retiring in their price row are left out of these three bands (4 on this check); they stay in the overall ranking. 60 models qualify; the 10 cheapest are shown.
| # | Model | Cheapest provider | Input | Output | Blended | Maker's own price |
|---|---|---|---|---|---|---|
| 1 | GPT OSS 20B | AkashML promo · 10 providers | USD 0.018 | USD 0.09 | USD 0.036 | — |
| 2 | Gemma 3 4B | Amazon Bedrock · 3 providers | USD 0.04 | USD 0.08 | USD 0.05 | — |
| 3 | Llama 3 8B Stheno V3.2 | Novita AI · 2 providers | USD 0.05 | USD 0.05 | USD 0.05 | — |
| 4 | Qwen 2.5 7B | SiliconFlow · 3 providers | USD 0.05 | USD 0.05 | USD 0.05 | — |
| 5 | Mistral-Small-24B-Instruct-2501 | DeepInfra · 2 providers | USD 0.05 | USD 0.08 | USD 0.0575 | — |
| 6 | Qwen3-8B | SiliconFlow · 2 providers | USD 0.06 | USD 0.06 | USD 0.06 | — |
| 7 | Llama 3.2 1B Instruct | Cloudflare Workers AI · 3 providers | USD 0.027 | USD 0.201 | USD 0.0705 | — |
| 8 | Gemma 3 12B | DeepInfra · 4 providers | USD 0.05 | USD 0.15 | USD 0.075 | — |
| 9 | IBM Granite 4.2 8B | CoreWeave Forge · 2 providers | USD 0.06 | USD 0.15 | USD 0.0825 | — |
| 10 | NVIDIA Nemotron 3.5 Lightning 30B A3B | DeepInfra · 5 providers | USD 0.06 | USD 0.16 | USD 0.085 | — |
What is the cheapest LLM API from the model maker itself?
The makers' own standard prices, on the same 3:1 blend: 154 models from 16 makers that sell an API. Lowest prompt-size tier; DeepSeek's rows are its peak-hour price. Models their maker lists as deprecated or retiring are left out. Across 98 models with both a maker price and two or more providers, a third-party host undercuts the maker on 29.
| # | Model | Maker's API | Input | Output | Blended | Checked |
|---|---|---|---|---|---|---|
| 1 | Qwen3.7 Flash | Alibaba Cloud qwen3.7-flash | USD 0.03 | USD 0.13 | USD 0.055 | 2026-11-14 |
| 2 | Nova Micro 1.0 | Amazon (Bedrock) Amazon Nova Micro | USD 0.035 | USD 0.14 | USD 0.06125 | 2026-11-14 |
| 3 | Command R7B (12-2024) | Cohere command-r7b-12-2024 | USD 0.0375 | USD 0.15 | USD 0.06562 | 2026-11-14 |
| 4 | GLM-4-32B-0414-128K | Z.ai GLM-4-32B-0414-128K | USD 0.10 | USD 0.10 | USD 0.10 | 2026-11-14 |
| 5 | Ministral 3 3B | Mistral AI ministral-3b-2512 | USD 0.10 | USD 0.10 | USD 0.10 | 2026-11-14 |
| 6 | Nova Lite 1.0 | Amazon (Bedrock) Amazon Nova Lite | USD 0.06 | USD 0.24 | USD 0.105 | 2026-11-14 |
| 7 | Granite 4.0 H Small | IBM watsonx.ai granite-4h-small | USD 0.0636 | USD 0.265 | USD 0.114 | 2026-11-15 |
| 8 | GLM-4.6V-FlashX | Z.ai GLM-4.6V-FlashX | USD 0.04 | USD 0.40 | USD 0.13 | 2026-11-14 |
| 9 | GPT-5 Nano | OpenAI gpt-5-nano | USD 0.05 | USD 0.40 | USD 0.1375 | 2026-11-14 |
| 10 | Qwen-Flash | Alibaba Cloud qwen-flash | USD 0.05 | USD 0.40 | USD 0.1375 | 2026-11-14 |
Where does BroadNet Inference stand?
BroadRouter is BroadNet's own site, so BroadNet is not in the ranking above. Here is where its direct prices would place on the same blend. None of its models is the cheapest per token.
| BroadNet model | Input | Output | Blended | Would rank |
|---|---|---|---|---|
| BroadNet Gemma 4 VL 26B | USD 0.10 | USD 0.22 | USD 0.13 | 30 of 198 · 29 cheaper |
| DeepSeek V4.1 Flash | USD 0.175 | USD 0.52 | USD 0.2612 | 59 of 198 · 58 cheaper |
| Qwen3.8 27B | USD 0.175 | USD 2.17 | USD 0.6737 | 105 of 198 · 104 cheaper |
| GLM 5.3 | USD 0.78 | USD 3.46 | USD 1.45 | 141 of 198 · 140 cheaper |
Where BroadNet can come out ahead is total cost for heavy, steady use: the unlimited DeepSeek V4.1 Flash plan costs $60 a month per session with no token meter. Whether that beats the cheapest per-token price depends on your volume; the flat-rate vs per-token calculator gives the break-even point. BroadNet also keeps zero data retention on every model; this page does not compare data policies, so check each provider's own terms before you send sensitive data.
What does the cheapest price leave out?
- Quality: a cheap small model can cost more per finished task if it needs retries. Price is one input to the choice.
- Cached input: agents resend long prefixes; a low cache price can matter more than the input price. Each model page lists it.
- Context, precision and region: the cheapest row may cap context lower, serve FP4 or apply to one region. The model page shows what the provider states.
- Tokens per word: tokenizers differ, so the same text costs a different number of tokens. See the tokenizer efficiency data and the token counter.
- Freshness: prices change. Each row carries the date we read it; the change log lists what moved.
All 428 models, every provider · LLM API price index (open data, CSV and JSON) · Download all prices · Price change log · LLM API cost calculator
Questions
What is the cheapest LLM API?
Mistral-Nemo-Instruct-2407 at DeepInfra: USD 0.029 per 1M input tokens and USD 0.03 per 1M output tokens, the lowest blended price (3 input tokens to 1 output) among the 198 models that pass the quality filter (two or more providers, standard tier, outliers set aside). Prices read from each provider's own pricing page, newest check 2026-11-15. The cheapest model is rarely the best one for a task, so price your workload before you pick.
What is the cheapest premium-priced model API?
GPT-5 (2025-08-07) at OpenAI, USD 1.25 input and USD 10 output per 1M tokens. "Premium-priced" means the maker's own output price is USD 10 or more per 1M tokens. The band is set by price alone, so it is not a list of each lab's newest model. Next: GPT-5.1 (USD 1.25 / USD 10 at OpenAI).
Is a model cheaper from a third-party host than from its maker?
Often, for open-weight models. Of 98 models with two or more providers and a maker price, a third-party host is cheaper than the maker on 29 on the 3:1 blend. Each model page shows the gap for that model.
Is BroadNet Inference the cheapest LLM API?
No, not per token. On the 3:1 blend, BroadNet's cheapest model, BroadNet Gemma 4 VL 26B, would rank 30 of 198. 29 models have a cheaper standard price. BroadNet competes on zero data retention and on an unlimited DeepSeek V4.1 Flash plan at $60 a month per session, which beats per-token billing only for heavy, steady use.
How is "cheapest" calculated here?
Each model's cheapest qualifying standard price, ranked on (3 × input + output) ÷ 4 in USD per 1M tokens. Only models with a standard price from 2 or more providers are ranked. Each is ranked on its cheapest standard, paid price. With 3 or more such prices, one under a third of the model's median price on input or output counts as an outlier unless another provider charges the same (within 5%): it stays on the model page but does not set the rank here. Batch, flex or spot, priority, fast and ultrafast tiers are shown on model pages but never ranked. Free rows are not ranked. Prices a provider calls promotional or discounted rank as published and are tagged "promo". DeepSeek rows use its peak-hour price. Cached-input prices are not in the ranking but are shown on each model page.
What is the cheapest LLM API from the model maker itself?
Qwen3.7 Flash on Alibaba Cloud's own API, at USD 0.03 input and USD 0.13 output per 1M tokens (checked 2026-11-14). Next: Nova Micro 1.0 from Amazon (Bedrock) at USD 0.035 / USD 0.14. Models their maker lists as deprecated or retiring are left out.
Read as MarkdownAsk ChatGPTAsk ClaudeAsk Perplexity
Sources
- AI21 Labs: pricing (checked 2026-11-15)
- AkashML: pricing (checked 2026-11-15)
- Alibaba Cloud Model Studio: pricing (checked 2026-11-14)
- Amazon Bedrock: pricing (checked 2026-11-14)
- Amazon Nova (Bedrock): pricing (checked 2026-11-14)
- Anthropic: pricing (checked 2026-11-14)
- Baidu: pricing (checked 2026-11-15)
- Baseten: pricing (checked 2026-11-14)
- Cerebras: pricing (checked 2026-11-14)
- Chutes: pricing (checked 2026-11-15)
- Cloudflare Workers AI: pricing (checked 2026-11-14)
- Cohere: pricing (checked 2026-11-14)
- CoreWeave Forge (formerly W&B Inference): pricing (checked 2026-11-15)
- Crusoe Cloud: pricing (checked 2026-11-15)
- DeepInfra: pricing (checked 2026-11-14)
- DeepSeek: pricing (checked 2026-11-14)
- DigitalOcean Gradient (Serverless Inference): pricing (checked 2026-11-14)
- Featherless: pricing (checked 2026-11-15)
- Fireworks AI: pricing (checked 2026-11-14)
- FriendliAI: pricing (checked 2026-11-15)
- GMI Cloud: pricing (checked 2026-11-15)
- Google (Gemini API): pricing (checked 2026-11-14)
- Google Cloud Vertex AI: pricing (checked 2026-11-14)
- Groq: pricing (checked 2026-11-14)
- Hyperbolic: pricing (checked 2026-11-14)
- IBM watsonx.ai: pricing (checked 2026-11-15)
- Microsoft Azure (Azure OpenAI + Foundry Models): pricing (checked 2026-11-14)
- MiniMax: pricing (checked 2026-11-14)
- Mistral AI: pricing (checked 2026-11-14)
- Moonshot AI (Kimi): pricing (checked 2026-11-14)
- Novita AI: pricing (checked 2026-11-14)
- Nscale: pricing (checked 2026-11-15)
- OpenAI: pricing (checked 2026-11-14)
- Parasail: pricing (checked 2026-11-14)
- Perplexity: pricing (checked 2026-11-15)
- SambaNova Cloud: pricing (checked 2026-11-14)
- SiliconFlow: pricing (checked 2026-11-14)
- Together AI: pricing (checked 2026-11-14)
- Venice: pricing (checked 2026-11-15)
- xAI: pricing (checked 2026-11-14)
- Z.ai: pricing (checked 2026-11-14)
- BroadNet Inference rate card, pricing.json (synced 2026-11-14)