AI API Pricing Comparison (October 2026): OpenAI vs Claude vs Gemini vs Qwen vs DeepSeek vs Kimi
AI API pricing in October 2026 spans 100x: budget models such as OpenAI gpt-6-luna and Claude Haiku 5.5 cost $0.10 per million input tokens and $0.50 per million output tokens, mid-tier workhorses such as gpt-6-sol and Claude Sonnet 5.5 cost $2 / $10, and flagships such as gpt-6-astra and Claude Fable 5.1 cost $10 / $50. Gemini, Qwen, DeepSeek, and Kimi undercut those at most tiers, with DeepSeek's off-peak flash model at $0.15 / $0.60 and Qwen 3.8 Max at $2 / $6. Two Gemini prices are temporary promotions that double on December 1, 2026 and January 1, 2027, so a comparison made today needs dates attached.
This page is for indie developers and small teams who pay per token and want one table instead of six pricing pages. Every number below comes from the provider's own pricing page, checked October 22, 2026. You get tier-by-tier tables, the discounts that change the ranking (cache, batch, off-peak), worked monthly costs for three workloads, a calculator, and how to choose.
AI API pricing table: all six providers at a glance
USD per million tokens (MTok), standard real-time pricing, short context.
| Provider | Flagship (in / out) | Mid-tier (in / out) | Budget (in / out) |
|---|---|---|---|
| OpenAI | gpt-6-astra $10 / $50 | gpt-6-sol, gpt-6.1-sol $2 / $10 | gpt-6-luna $0.10 / $0.50 |
| Anthropic Claude | Fable 5.1 $10 / $50; Opus 5.5 $4 / $20 | Sonnet 5.5 $2 / $10 | Haiku 5.5 $0.10 / $0.50 (prompts up to 100K) |
| Google Gemini | Gemini 4 Argon $2 / $10 until Nov 30, 2026, then $4 / $20 | Gemini 3.8 Flash $0.75 / $3.75 until Dec 31, 2026, then $1.50 / $7.50 | Gemini 3.1 Flash-Lite $0.25 / $1.50 |
| Alibaba Qwen (international) | qwen3.8-max $2 / $6 | qwen3.7-plus $0.40 / $1.60 | qwen3.8-flash $0.15 / $0.47 |
| DeepSeek | deepseek-v4-pro $0.66 / $1.98 off-peak ($1.32 / $3.96 peak) | n/a | deepseek-flash $0.15 / $0.60 off-peak ($0.30 / $1.20 peak) |
| Moonshot Kimi | kimi-k3 $3 / $15 | kimi-k2.6 $0.95 / $4 | n/a |
Sources linked in the first column; all checked October 22, 2026. "Flagship," "mid-tier," and "budget" are our grouping by each provider's own positioning, not a quality benchmark. We did not run the models; this is a price comparison only.
What jumps out
- OpenAI and Anthropic now price their tiers identically at the top ($10 / $50), middle ($2 / $10), and bottom ($0.10 / $0.50). The choice between them is about quality on your task, not price.
- Output costs 5x input for OpenAI, Claude, Gemini 4 Argon, and Kimi K3. Qwen 3.8 Max (3x) and DeepSeek (3x to 4x) are flatter, which favors long answers.
- Gemini's current prices are promotional. Gemini 4 Argon doubles on December 1, 2026; Gemini 3.8 Flash doubles on January 1, 2027. Budget at the later price if your launch runs past those dates.
- DeepSeek has time-of-day pricing. Peak hours (01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday) cost twice the off-peak rate.
Discounts that change the ranking
List price is rarely what you pay. These are the official discount mechanics per provider, checked October 22, 2026.
| Provider | Cached input | Batch / async | Other |
|---|---|---|---|
| OpenAI | Cache reads 5% to 10% of input (gpt-6.1-sol $0.10, gpt-6-sol $0.20); cache writes 1.25x (30 min) or 2.5x (12 h) | Batch and Flex: 50% off | Fast processing: 2x; regional data residency endpoints: +10%; prompts over 272K tokens: 2x input, 1.5x output |
| Claude | Reads 2.5% to 10% of input; writes 1.25x (5 min) or 2x (1 h) | Batch: 50% off | US-only inference: 1.1x |
| Gemini | Argon $0.10, 3.8 Flash $0.075 per MTok | Batch: 50% off | Free tier on Flash models (data used to improve products) |
| Qwen | Cache hits 10% of input | Batch: 50% off | Limited-time discounts on some models |
| DeepSeek | Cache hits from $0.003 per MTok off-peak | n/a | Off-peak is half of peak |
| Kimi | K3 cached input $0.30 | Batch: 60% of standard price | $5 voucher after $5 cumulative recharge |
Two practical consequences:
- For workloads that can wait, batch beats switching providers. Claude Sonnet 5.5 in batch is $1 / $5, the same as Gemini 4 Argon's batch price until November 30.
- For agents that resend long context, cache price matters more than input price. A 50K-token prefix read from cache costs $0.005 on Sonnet 5.5 or gpt-6.1-sol, versus $0.10 uncached. See our prompt caching ROI guide.
Worked monthly costs for three workloads
All at list prices, standard tier, no caching, October 22, 2026.
Workload A: support classifier. 1 million requests, 800 input and 50 output tokens each (800M in, 50M out).
| Model | Monthly cost |
|---|---|
| gpt-6-luna | $105 |
| Claude Haiku 5.5 | $105 |
| qwen3.8-flash | $143.50 |
| deepseek-flash (off-peak) | $150 |
| Gemini 3.1 Flash-Lite | $275 |
Workload B: RAG answers. 200,000 requests, 4,000 input and 500 output tokens each (800M in, 100M out).
| Model | Monthly cost |
|---|---|
| qwen3.7-plus | $480 |
| Gemini 3.8 Flash (promo price) | $975 |
| Kimi K2.6 | $1,160 |
| gpt-6-sol / Claude Sonnet 5.5 | $2,600 |
Workload C: coding agent. 5,000 sessions, 200,000 input and 20,000 output tokens each (1,000M in, 100M out).
| Model | Monthly cost |
|---|---|
| deepseek-v4-pro (off-peak) | $858 |
| qwen3.8-max | $2,600 |
| Gemini 4 Argon (promo, until Nov 30) | $3,000 |
| Kimi K3 | $4,500 |
| Claude Opus 5.5 | $6,000 |
| gpt-6-astra / Claude Fable 5.1 | $15,000 |
Coding agents usually hit 70% to 90% cache rates, which cuts the input share of these figures by most of its value. Run the calculator below with your real cache rate.
AI API cost calculator
# List prices per 1M tokens (input, output), checked Oct 22, 2026 on each provider's pricing page
PRICES = {
"gpt-6-astra": (10.00, 50.00), "gpt-6-sol": (2.00, 10.00), "gpt-6-luna": (0.10, 0.50),
"claude-fable-5-1": (10.00, 50.00), "claude-opus-5-5": (4.00, 20.00),
"claude-sonnet-5-5": (2.00, 10.00), "claude-haiku-5-5": (0.10, 0.50),
"gemini-4-argon": (2.00, 10.00), # $4 / $20 from Dec 1, 2026
"gemini-3.8-flash": (0.75, 3.75), # $1.50 / $7.50 from Jan 1, 2027
"gemini-3.1-flash-lite": (0.25, 1.50),
"qwen3.8-max": (2.00, 6.00), "qwen3.7-plus": (0.40, 1.60), "qwen3.8-flash": (0.15, 0.47),
"deepseek-v4-pro-offpeak": (0.66, 1.98), "deepseek-flash-offpeak": (0.15, 0.60),
"kimi-k3": (3.00, 15.00), "kimi-k2.6": (0.95, 4.00),
}
def monthly(model, requests, tokens_in, tokens_out, batch=False):
pin, pout = PRICES[model]
cost = requests * (tokens_in * pin + tokens_out * pout) / 1_000_000
return cost * (0.5 if batch else 1.0)
for m in ["gpt-6-luna", "claude-haiku-5-5", "qwen3.8-flash", "deepseek-flash-offpeak"]:
print(f"{m:26s} ${monthly(m, 1_000_000, 800, 50):,.2f}")
To check what a request really cost, read the usage field the API returns:
curl -s https://api.openai.com/v1/chat/completions \
-H "Authorization: Bearer $OPENAI_API_KEY" -H "Content-Type: application/json" \
-d '{"model":"gpt-6-luna","messages":[{"role":"user","content":"Say hi"}]}' \
| python3 -c "import sys,json; print(json.load(sys.stdin)['usage'])"
// Node: same check with the OpenAI SDK
import OpenAI from "openai";
const client = new OpenAI();
const r = await client.chat.completions.create({
model: "gpt-6-luna",
messages: [{ role: "user", content: "Say hi" }],
});
console.log(r.usage); // { prompt_tokens, completion_tokens, ... }
How to choose without overpaying
- Start from the job, not the cheapest cell. Classification and extraction rarely need a flagship; a $0.10 model at 1M requests a month costs about $100.
- Compare at your input/output ratio. Long-answer workloads favor Qwen 3.8 Max and DeepSeek because their output multiplier is lower.
- Check the date column. Gemini's promo prices and DeepSeek's peak hours both move your bill without any code change.
- Count tokens per model. Tokenizers differ; Anthropic says its Claude 4.7+ tokenizer produces about 30% more tokens for the same text than older Claude models.
- Route by step. Use a cheap model for routing and a strong one only for the hard step. Our model routing policy guide walks through this.
Comparing through one key: Flatkey prices
Flatkey serves models from all six providers through one OpenAI- and Anthropic-compatible API key with prepaid credit. On October 22, 2026, the Flatkey model directory listed:
| Model | Provider list (in / out) | Flatkey (in / out) |
|---|---|---|
| Claude Sonnet 5.5 | $2 / $10 | $1.60 / $8 |
| gpt-6-sol | $2 / $10 | $1.20 / $6 |
| Gemini 4 Argon | $2 / $10 (promo) | $1.60 / $8 |
| qwen3.8-max | $2 / $6 | $1.60 / $4.80 |
| Kimi K3 | $3 / $15 | $1.80 / $9 |
| Gemini 3.8 Flash | $0.75 / $3.75 (promo) | $1.20 / $6 |
Note the last row: while Google's 3.8 Flash promotion runs, calling Google directly is cheaper than Flatkey. A gateway is worth it when you want one bill, fallbacks, and one SDK across vendors; see AI gateway pricing for what gateways themselves charge.
Provider deep dives
- Anthropic API pricing: every Claude model, caching, batch, and spend caps
- Gemini API pricing: promo dates and free tier
- DeepSeek API pricing: peak and off-peak rates
- Qwen API guide: regions and OpenAI-compatible setup
- Kimi API pricing and K3 guide
For a head-to-head on the two most common shortlist picks, see Gemini vs Claude: which AI should you use?
Prices are only half the story; our AI cost optimization guide shows how routing, prompt caching, and batch lower the bill.
FAQ
What is the cheapest AI API in October 2026?
Among the six providers here, the lowest list input price is $0.10 per million tokens (gpt-6-luna and Claude Haiku 5.5), and qwen3.8-flash has the lowest output price at $0.47. DeepSeek's cache hits start at $0.003 per million tokens off-peak.
How is AI API pricing calculated?
You pay per token: input tokens times the input price plus output tokens times the output price, each per million tokens. Cached input, batch processing, long-context surcharges, and tool calls such as web search adjust the total.
Is Claude more expensive than OpenAI?
Not at list price in October 2026. Both price their three tiers at $10 / $50, $2 / $10, and $0.10 / $0.50. Claude Opus 5.5 ($4 / $20) has no direct OpenAI price match.
Which AI API has the cheapest output tokens?
qwen3.8-flash at $0.47 per million output tokens, followed by gpt-6-luna and Claude Haiku 5.5 at $0.50 and deepseek-flash off-peak at $0.60.
Do these prices include prompt caching?
No. The main table shows standard, uncached prices. Cached input is 2.5% to 10% of the input price at OpenAI and Anthropic, 10% at Qwen, and listed separately for Gemini, DeepSeek, and Kimi.
How often do AI API prices change?
Often. Two Gemini prices change on fixed dates in the next three months, and new model releases usually come with new price points. Recheck the official pages before committing a budget.
Sources
- OpenAI API pricing, Anthropic pricing, Gemini API pricing, Alibaba Model Studio pricing, DeepSeek pricing, Kimi pricing, all checked October 22, 2026
- Flatkey model directory, checked October 22, 2026
