Unboxing the AI Stack
L02 · Language Models · Ranking · Top 19

Cheapest LLM APIs per 1M Output Tokens (October 2026)

Nineteen flagship and low-cost language models from ten vendors, ranked by list price per 1M output tokens on each vendor’s own pricing page. Alibaba’s Qwen3.8-Flash is cheapest at $0.47; two flagships list $50.

Output price, eight cheapest models USD per 1M tokens

Qwen3.8-Flash$0.47
Claude Haiku 5.5$0.50
GPT-6 Luna$0.50
GLM-5.3-Flash$0.50
Mistral Small 4$0.60
DeepSeek Flash$1.20
MiniMax-M3$1.20
Gemini 3.1 Flash-Lite$1.50

Vendor-stated list prices, standard tier, as of Oct 13, 2026. Bars are relative to the highest price shown ($1.50).

Chart: list price per 1M output tokens for the eight cheapest models in the table, from Qwen3.8-Flash ($0.47) to Gemini 3.1 Flash-Lite ($1.50).

As of All prices vendor-statedUSD per 1M tokens · standard tier · ties share a rank

Cheapest LLM APIs per 1M Output Tokens (October 2026), list prices in USD per 1M tokens, as of 2026-10-13
Rank Model Vendor Tier Output ($ / 1M tokens) Input ($ / 1M tokens) Price basis Figure
1 Qwen3.8-Flash7qwen3.8-flash Alibaba Cloud Low-cost $0.47 $0.15 Singapore, prompts ≤1M Vendor-stated
2 Claude Haiku 5.52 Anthropic Low-cost $0.50 $0.10 Prompts ≤100K Vendor-stated
2 GPT-6 Luna1gpt-6-luna OpenAI Low-cost $0.50 $0.10 Short context Vendor-stated
2 GLM-5.3-Flash9 Z.ai Low-cost $0.50 $0.15 Single price Vendor-stated
5 Mistral Small 411 Mistral AI Low-cost $0.60 $0.15 List price Vendor-stated
6 DeepSeek Flash6deepseek-flash (DeepSeek-V4.1-Flash) DeepSeek Low-cost $1.20 $0.30 Peak hours Vendor-stated
6 MiniMax-M310MiniMax-M3 MiniMax Flagship $1.20 $0.30 Prompts ≤512K Vendor-stated
8 Gemini 3.1 Flash-Lite3gemini-3.1-flash-lite Google Low-cost $1.50 $0.25 Text input Vendor-stated
9 Grok 4.34grok-4.3 xAI Low-cost $2.50 $1.25 Prompts <200K Vendor-stated
10 DeepSeek V4 Pro6deepseek-v4-pro (DeepSeek-V4-Pro-0813) DeepSeek Flagship $3.96 $1.32 Peak hours Vendor-stated
11 Kimi K2.68kimi-k2.6 Moonshot AI Low-cost $4.00 $0.95 Cache-miss input Vendor-stated
12 Mistral Large 411 Mistral AI Flagship $4.18 $1.36 List price (not sale) Vendor-stated
13 GLM-5.39 Z.ai Flagship $4.40 $1.40 Single price Vendor-stated
14 Qwen3.8-Max7qwen3.8-max Alibaba Cloud Flagship $6.00 $2.00 Singapore, prompts ≤1M Vendor-stated
14 Grok 4.74grok-4.7 xAI Flagship $6.00 $2.00 Prompts <200K Vendor-stated
16 Gemini 3.1 Pro Preview3gemini-3.1-pro-preview Google Flagship $12.00 $2.00 Prompts ≤200K Vendor-stated
17 Kimi K38kimi-k3 Moonshot AI Flagship $15.00 $3.00 Cache-miss input Vendor-stated
18 Claude Fable 5.12 Anthropic Flagship $50.00 $10.00 Single price Vendor-stated
18 GPT-6 Astra1gpt-6-astra OpenAI Flagship $50.00 $10.00 Short context Vendor-stated

Metric: the list price per 1M output tokens in US dollars, as stated on each vendor's own API pricing page on Oct 13, 2026, with the input price (cache miss) beside it; no blended price is computed. One rule for every row: the standard tier (not batch, flex, priority or fast modes); the shortest-context price where the price depends on prompt length; DeepSeek's peak price (its off-peak rates are half); the list price, not a limited-time sale (Mistral Large 4 is on sale at $0.68 input and $2.09 output; MiniMax labels its 50% cut on M3 permanent, so the reduced price counts); Alibaba's Singapore (International) deployment.671011 Free tiers and free quotas aren't rows. Rows: two per vendor from its current lineup, its top-tier model and its low-cost tier (Flash, Flash-Lite, Haiku, Luna, Small); where a vendor sells two current low-cost models, the cheaper one (Gemini 3.1 Flash-Lite, not 3.5 Flash-Lite at $2.50 output).3 For Mistral the top tier is Large 4, listed first in its "Flagship models" table, although Medium 3.5 lists a higher price ($7.50 output).11 xAI and Moonshot AI sell no separate low-cost tier, so their second row is the cheapest other current general model on the price list (Grok 4.3, Kimi K2.6); xAI's grok-build-0.1, named for its Grok Build product, is left out.48 MiniMax lists one current model. Older or smaller models with lower list prices are left out: Google's Gemini 2.5 Flash-Lite ($0.40 output), Alibaba's qwen3.7-flash ($0.13 output for requests up to 32K tokens) and Mistral's Ministral 3 3B ($0.10).3711 So are mid-tier models such as GPT-6.1 Sol, Claude Opus 5.5 and Sonnet 5.5, and Gemini 3.8 Flash, whose price Google's page says doubles on January 1, 2027.123 Tokens aren't a common unit: each vendor's tokenizer splits text differently, so a lower price per token isn't automatically a lower cost for the same job. Anthropic says the tokenizer in Claude 4.7 and later models "produces approximately 30% more tokens for the same text" than its earlier one; Moonshot puts a typical English token at 3 to 4 characters, and MiniMax estimates 750 English words at about 1,000 tokens.2810 Reasoning models also bill their thinking tokens as output (Google's page says so explicitly), so the output bill depends on how much a model thinks.3 Figure labels: how they work.

The top 10

The ten cheapest rows by output price, from $0.47 to $3.96 per 1M tokens. Ties share a rank, and tied rows are ordered by input price. At the other end of the table, OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1 both list $50 per 1M output tokens, more than 100 times the price of the cheapest row (our ratio of the two list prices).12

#1

Qwen3.8-Flash (Alibaba Cloud)

$0.47 / 1M output

Alibaba Cloud's Model Studio lists qwen3.8-flash at $0.15 per 1M input tokens and $0.47 per 1M output tokens in its Singapore (International) deployment, one price for any request up to 1M input tokens.7 It qualifies for Model Studio's context-caching discount, and Alibaba offers a free quota of 1 million tokens, valid for 90 days and in Singapore only, which this ranking doesn't count.7 Alibaba's previous qwen3.7-flash lists less, $0.13 per 1M output tokens for requests up to 32K tokens, but the table takes each vendor's current lineup.7

#2

Claude Haiku 5.5 (Anthropic)

$0.50 / 1M output

Anthropic describes Haiku 5.5 as its model for high-volume, latency-sensitive tasks such as classification, extraction and routing.2 It's the only model in Anthropic's main price table whose price depends on prompt length: $0.10 input and $0.50 output per 1M tokens for prompts up to 100,000 tokens, and $0.50 and $2.50 above that. Cache hits cost $0.01 per 1M tokens.2 Anthropic also notes that Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text, which matters when comparing its prices with other vendors'.2

#2

GPT-6 Luna (OpenAI)

$0.50 / 1M output

gpt-6-luna is the cheapest of the three models in OpenAI's "Flagship models" price table, at $0.10 input and $0.50 output per 1M tokens for short-context requests; long-context requests cost $0.20 and $0.75.1 OpenAI's Batch and Flex tiers halve the short-context rates to $0.05 and $0.25, and its Fast mode doubles them to $0.20 and $1.00. Cached input costs $0.01 per 1M tokens.1

#2

GLM-5.3-Flash (Z.ai)

$0.50 / 1M output

Z.ai lists GLM-5.3-Flash in its "Latest models" table at $0.15 input, $0.03 cached input and $0.50 output per 1M tokens, with cached-input storage marked as limited-time free.9 Its sibling GLM-5.3-FlashX costs $0.37 input and $1.25 output.9 It ties OpenAI's and Anthropic's low-cost models on output price but lists $0.15 for input, against their $0.10.12

#5

Mistral Small 4 (Mistral AI)

$0.60 / 1M output

Mistral describes Small 4 as a hybrid model that unifies instruct, reasoning and coding in a single efficient model, and lists it under the Apache 2.0 license.12 Its API price is $0.15 input, $0.015 cached input and $0.60 output per 1M tokens.11 Mistral's own pricing page points cost-sensitive projects to Mistral Small.13 The smaller Ministral 3 models list less, down to $0.10 for both input and output on Ministral 3 3B.11

#6

DeepSeek Flash (DeepSeek)

$1.20 / 1M output

The deepseek-flash model name now serves DeepSeek-V4.1-Flash; the legacy deepseek-v4-flash names still work but are served and billed as Flash.6 At peak hours it costs $0.30 input (cache miss) and $1.20 output per 1M tokens, and off-peak rates are half. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays.6 DeepSeek lists a 1M context length and up to 384K output tokens, and Flash accepts images, which V4 Pro doesn't.6

#6

MiniMax-M3 (MiniMax)

$1.20 / 1M output

MiniMax lists M3 at $0.30 input and $1.20 output per 1M tokens for requests up to 512K input tokens, down from $0.60 and $2.40 in a cut its pricing page labels "Permanent 50% off"; above 512K input tokens the rates are $0.60 and $2.40.10 Its Priority tier costs 1.5 times the standard rate.10 M3 is the only current model in MiniMax's standard price table, which makes it the cheapest top-tier row in this ranking.10

#8

Gemini 3.1 Flash-Lite (Google)

$1.50 / 1M output

Google describes Gemini 3.1 Flash-Lite as a cost-efficient model optimized for high-volume agentic tasks, translation and simple data processing.3 Paid-tier input costs $0.25 per 1M tokens for text, image or video and $0.50 for audio, and output, including thinking tokens, costs $1.50.3 Google's newer Gemini 3.5 Flash-Lite lists $2.50 per 1M output tokens, and the older Gemini 2.5 Flash-Lite $0.40.3

#9

Grok 4.3 (xAI)

$2.50 / 1M output

xAI calls Grok 4.7 its flagship model, but its price list still carries Grok 4.3 at $1.25 input and $2.50 output per 1M tokens, the same rates as its Grok 4.20 models.45 Once a prompt reaches 200K tokens, xAI bills every token in the request at the long-context rates, $2.50 and $5.00.4 Grok 4.3 has a 1M-token context window, against 500K for Grok 4.7, and requests sent to xAI's US regional endpoint cost 1.1 times the global rates.4

#10

DeepSeek V4 Pro (DeepSeek)

$3.96 / 1M output

DeepSeek's larger model, served as DeepSeek-V4-Pro-0813, costs $1.32 input (cache miss) and $3.96 output per 1M tokens at peak hours, and $0.66 and $1.98 off-peak.6 It shares Flash's 1M context length but doesn't accept images, and its concurrency limit is 500 against Flash's 2,500.6 After MiniMax-M3, it's the cheapest top-tier model in the table, just under Moonshot AI's Kimi K2.6 ($4.00) and Mistral Large 4 ($4.18 at list price).811

Sources

  1. OpenAI, Pricing (OpenAI API docs), retrieved 2026-10-13.
  2. Anthropic, Pricing (Claude API docs), retrieved 2026-10-13.
  3. Google, Gemini Developer API pricing, last updated 2026-10-09, retrieved 2026-10-13.
  4. xAI, Pricing (xAI API docs), last updated October 10, 2026, retrieved 2026-10-13.
  5. xAI, Models (xAI API docs), last updated September 21, 2026, retrieved 2026-10-13.
  6. DeepSeek, Models & Pricing (DeepSeek API docs), retrieved 2026-10-13.
  7. Alibaba Cloud, Model Studio: Model inference pricing, last updated Oct 13, 2026, retrieved 2026-10-13.
  8. Moonshot AI, Model Inference Pricing Explanation (Kimi API Platform), retrieved 2026-10-13.
  9. Z.ai, Pricing (Z.ai developer docs), retrieved 2026-10-13.
  10. MiniMax, Pay as You Go (MiniMax API docs), retrieved 2026-10-13.
  11. Mistral AI, API pricing (Mistral docs), retrieved 2026-10-13.
  12. Mistral AI, Models Overview (Mistral docs), retrieved 2026-10-13.
  13. Mistral AI, Pricing, FAQ, retrieved 2026-10-13.