Cheapest LLM API: AI model prices compared
Claude Haiku 5.5 and GPT-6 Luna are the cheapest LLM APIs we track, at $0.20 per 1M tokens (blended). For 1M-token context from the model's own maker, DeepSeek Flash is the best value (MiniMax M3.1 Flash costs less than its peak price, but as a Together AI reseller price). Want quality instead? See the tier lists.
Last verified: · On 2 Nov 2026 the input and output prices of every model in the table except Claude Fable 5.1 and Claude Mythos 5.1 were compared with the official pricing pages (OpenAI, Anthropic, Google, xAI, DeepSeek, Mistral) or, for the three rows resold by Together AI (Kimi K3, GLM-5.3, MiniMax M3.1 Flash), with Together AI's model pages, and matched the table. Cached-input rates were also compared for OpenAI, Anthropic, Google, DeepSeek and Mistral, plus Google's promo end dates and DeepSeek's off-peak hours. Long-context surcharges, batch discounts, Grok's cached rate and the Together AI cached rates were not re-read that day. Claude Fable 5.1 and Claude Mythos 5.1 were added on 3 Nov 2026 from Anthropic's pricing and model pages (input, output, cached and batch compared that day).
Top 3 value picks
We pick these three by hand from the table, one per need: the most tokens per $1 with a verified 1M context, the cheapest model from a frontier lab, and the cheapest flagship. Each shows its blended price (3 input : 1 output tokens) and tokens per $1.
DeepSeek Flash (V4.1)
$0.525 /1M blended · 1,904,762 tokens per $1
Best tokens per dollar with a verified 1M context among models sold directly by their maker (MiniMax M3.1 Flash costs less than its peak price, but as a Together AI reseller price). Off-peak hours make it about 3.8M tokens per $1.
GPT-6 Luna
$0.20 /1M blended · 5,000,000 tokens per $1
The cheapest model from a frontier lab at a $0.20 blend, ties with Claude Haiku 5.5 (Haiku's price applies only to prompts up to 100K).
Grok 4.7
$3.00 /1M blended · 333,333 tokens per $1
The cheapest flagship-tier model sold directly by its maker, at a $3.00 blend, with a 500K context (GLM-5.3 costs less, but as a Together AI reseller price).
Cheapest LLM API prices per 1M tokens
| Claude Haiku 5.5Anthropic | $0.10 | $0.50 | $0.01 | $0.20 | 5,000,000 | Not verified | Official |
|---|---|---|---|---|---|---|---|
| GPT-6 LunaOpenAI | $0.10 | $0.50 | $0.01 | $0.20 | 5,000,000 | Not verified | Official |
| MiniMax M3.1 Flash (Together AI)MiniMax AI via Together AI (reseller price)Kimi, MiniMax, GLM, DeepSeek compared | $0.15 | $0.60 | $0.015 | $0.263 | 3,809,524 | 1M | Together AI |
| Mistral Small 4Mistral AI | $0.15 | $0.60 | $0.015 | $0.263 | 3,809,524 | 256K | Official |
| DeepSeek Flash (V4.1)DeepSeekKimi, MiniMax, GLM, DeepSeek compared | $0.30 | $1.20 | $0.006 | $0.525 | 1,904,762 | 1M | Official |
| Mistral Large 4Mistral AI | $0.70 | $2.10 | $0.07 | $1.05 | 952,381 | 1M | Official |
| Gemini 3.8 FlashGoogle Promo until 31 Dec 2026 | $0.75 | $3.75 | $0.075 | $1.50 | 666,667 | Not verified | Official |
| DeepSeek V4 ProDeepSeekKimi, MiniMax, GLM, DeepSeek compared | $1.32 | $3.96 | $0.044 | $1.98 | 505,051 | 1M | Official |
| GLM-5.3 (Together AI)ZAI via Together AI (reseller price)Kimi, MiniMax, GLM, DeepSeek compared | $1.40 | $4.40 | $0.26 | $2.15 | 465,116 | 1M | Together AI |
| Grok 4.7xAI | $2.00 | $6.00 | $0.50 | $3.00 | 333,333 | 500K | Official |
| Claude Sonnet 5.5Anthropic | $2.00 | $10.00 | $0.10 | $4.00 | 250,000 | 1M | Official |
| GPT-6.1 SolOpenAI | $2.00 | $10.00 | $0.10 | $4.00 | 250,000 | Not verified | Official |
| Gemini 4 ArgonGoogle Promo until 30 Nov 2026 | $2.00 | $10.00 | $0.10 | $4.00 | 250,000 | Not verified | Official |
| Kimi K3 (Together AI)Moonshot AI via Together AI (reseller price)Kimi, MiniMax, GLM, DeepSeek compared | $3.00 | $15.00 | $0.30 | $6.00 | 166,667 | 1M | Together AI |
| Claude Opus 5.5Anthropic | $4.00 | $20.00 | $0.20 | $8.00 | 125,000 | 1M | Official |
| Claude Fable 5.1Anthropic | $10.00 | $50.00 | $0.25 | $20.00 | 50,000 | 1M | Official |
| Claude Mythos 5.1Anthropic Limited accessVerified organizations only | $10.00 | $50.00 | $0.25 | $20.00 | 50,000 | 1M | Official |
Cheapest LLM API by provider
The cheapest model from each vendor in the table, by blended cost per 1M tokens (3 input : 1 output). The cheapest GPT model here is GPT-6 Luna.
- Anthropic: Claude Haiku 5.5, $0.20 per 1M tokens blended ($0.10 input / $0.50 output)
- OpenAI: GPT-6 Luna, $0.20 per 1M tokens blended ($0.10 input / $0.50 output)
- MiniMax AI via Together AI (reseller price): MiniMax M3.1 Flash (Together AI), $0.263 per 1M tokens blended ($0.15 input / $0.60 output) · Together AI's price, not MiniMax's own. Together's model page shows a single price and a 1M context.
- Mistral AI: Mistral Small 4, $0.263 per 1M tokens blended ($0.15 input / $0.60 output)
- DeepSeek: DeepSeek Flash (V4.1), $0.525 per 1M tokens blended ($0.30 input / $1.20 output)
- Google: Gemini 3.8 Flash, $1.50 per 1M tokens blended ($0.75 input / $3.75 output)
- ZAI via Together AI (reseller price): GLM-5.3 (Together AI), $2.15 per 1M tokens blended ($1.40 input / $4.40 output) · Together AI's price, not ZAI's own. Together's model page shows a single price and a 1M context.
- xAI: Grok 4.7, $3.00 per 1M tokens blended ($2.00 input / $6.00 output)
- Moonshot AI via Together AI (reseller price): Kimi K3 (Together AI), $6.00 per 1M tokens blended ($3.00 input / $15.00 output) · Together AI's price, not Moonshot AI's own. Together's model page shows a single price and a 1.05M context.
Monthly cost calculator
Enter your workload. Costs update instantly and work offline, because all prices are embedded in this page. It leaves out limited-access models (Claude Mythos 5.1).
Cheapest by job
Coding agents and code review
Claude Sonnet 5.5 for quality per dollar; Gemini 3.8 Flash or DeepSeek Flash for budget loops.
Chat and support bots
Small models with cheap cached input: GPT-6 Luna or Claude Haiku 5.5.
Long-document RAG (100K+ tokens)
Pick models with verified 1M context and no surcharge at your prompt size: DeepSeek Flash or Claude Sonnet 5.5. Haiku 5.5 costs 5× more above 100K.
Bulk extraction and classification
Run GPT-6 Luna or Claude Haiku 5.5 through the Batch API at 50% off, or DeepSeek Flash off-peak.
Hidden costs to watch
- Promo prices end:
- Gemini 3.8 Flash: promo price until , then $1.50 / $7.50.
- Gemini 4 Argon: promo price until , then $4.00 / $20.00.
- Long-context surcharges: OpenAI above 272K input, xAI at 200K+, and Claude Haiku 5.5 above 100K all charge more for the whole request.
- Tokenizers: Anthropic says Claude 4.7 and later models produce about 30% more tokens for the same text.
- Peak hours: DeepSeek doubles its price during peak hours (01:00–04:00 and 06:00–10:00 UTC, weekdays).
- Extras: search grounding, tool calls, cache writes and data-residency uplifts (often +10%) are not in the table.
Methodology
Every price is copied from the vendor's official pricing page or, for rows labelled as reseller prices, from Together AI's model page (linked in each row), and dated. Blended = (3 × input + 1 × output) ÷ 4. Tokens per $1 = 1,000,000 ÷ blended. Anything we can't verify is marked "Not verified" or left out. Llama, Kimi, GLM and MiniMax rows are priced on Together AI, a major host, and labelled as reseller prices: they come from Together AI's model pages, not from Meta, Moonshot AI, ZAI or MiniMax. Prices are editorial data, never sponsored. The raw data is in our llms.txt.
Changelog
- Added Claude Fable 5.1 ($10 input / $50 output, cached $0.25) and Claude Mythos 5.1 (same price) from Anthropic's pricing page and model pages, as unranked rows. Mythos 5.1 is labelled Limited access: Anthropic offers it only to organizations verified through its programs, so most readers cannot use it. It is left out of the calculator.
- Weekly check: no price changes. Removed Llama 3.3 70B (Together AI): Together AI's model page now says it is not available on its serverless API, and its pricing page no longer lists a serverless price (the old $1.04 / $1.04 can't be verified). Still no serverless Llama 4 price listed.
- Added Mistral Large 4 ($0.70 input / $2.10 output) and Mistral Small 4 ($0.15 / $0.60) from Mistral's pricing docs. Mistral was left out on 15 Oct because no per-model price had been published; its docs now list them. Mistral's other models are not in this table.
- Re-read the DeepSeek pricing page and the Together AI pages for Kimi K3, GLM-5.3 and MiniMax M3.1 Flash: input, output, cached and context figures unchanged. A new page compares them with Moonshot's, Z.ai's and MiniMax's own API prices.
- Added Kimi K3, GLM-5.3 and MiniMax M3.1 Flash at Together AI's reseller prices (not the makers' own), from Together AI's model pages. Llama 3.3 70B was already listed as a Together AI price.
- Weekly check against the official pricing pages: no price changes found.
- Weekly check: no price changes.
- Page launched with 11 models. Mistral dropped because no per-model API price is published on mistral.ai/pricing.
FAQ
What is the cheapest LLM API right now?
By blended cost (3 input : 1 output) among the models in the table, Claude Haiku 5.5 and GPT-6 Luna tie at $0.20 per 1M tokens as of 3 Nov 2026. DeepSeek Flash is half price off-peak.
How is the blended price calculated?
Blended = (3 × input price + 1 × output price) ÷ 4, a typical 3:1 input:output mix. Tokens per $1 = 1,000,000 ÷ blended price.
How often are these prices updated?
We check every official pricing page weekly, and on known promo end dates. Last update: 3 Nov 2026.
Which Mistral models are in the table?
Mistral Large 4 ($0.70 input / $2.10 output per 1M tokens) and Mistral Small 4 ($0.15 / $0.60), from Mistral's pricing docs. We left Mistral out on 15 Oct while no per-model price could be verified; its docs now list one. Mistral's other models, such as the smaller Ministral 3 models, Codestral and the OCR and audio models, are not in this table.
Can I use Claude Mythos 5.1?
Probably not. Claude Mythos 5.1 is priced at $10.00 input / $50.00 output per 1M tokens, the same as Claude Fable 5.1, but Anthropic offers it only to organizations it has verified through programs such as its Cyber Verification Program. It is labelled Limited access in the table, is not a pick, and is left out of the calculator. Claude Fable 5.1 is the version on the Claude API for everyone.