OpenAI API alternatives: OpenRouter prices, Sep 2026

OpenAI’s gpt-oss-20b has the lowest listed input price per million tokens from OpenRouter’s public model API on 28 September 2026: $0.018. It ranks first of 318 paid models from 13 labs, ahead of Mistral Nemo at $0.019 and DeepSeek V4 Flash 0731 at $0.021.¹ So nothing undercuts OpenAI at the bottom of the list. The cheaper alternatives sit higher up. GPT-6 Sol lists at $2.00, while Grok 4.7 lists at $1.60, Mistral Medium 3.5 at $1.50 and DeepSeek V4 Pro at $0.95526.¹

TL;DR

• OpenAI owns the floor on input price. gpt-oss-20b lists at $0.018, one thousandth of a dollar below Mistral Nemo, but Nemo lists output at $0.03 against OpenAI’s $0.09.¹
• Anthropic matches OpenAI at the top two tiers. Claude Fable 5.1 and GPT-6 Astra both list at $10.00 input and $50.00 output.¹ Claude Sonnet 5 and GPT-6 Sol both list at $2.00 and $10.00.¹
• Below GPT-6 Sol the field opens up. 239 of the 318 paid models list input below $2.00.¹
• The last nine rows of the ranking all belong to OpenAI, ending with o1-pro at $150.00 input and $600.00 output.¹
• OpenRouter matched the lab’s own page for every GPT-6, Claude, Gemini and Mistral price we checked. It listed Grok 4.7 at $1.60 input against $2.00 on xAI’s page, and no DeepSeek model we checked matched DeepSeek’s own table.^1,2,3

How this ranking was built

Glamdring Research ranked every paid model from 13 AI labs on one field, as part of our model infrastructure coverage: the listed input price in OpenRouter’s public model API, captured on 28 September 2026. The field is pricing.prompt, which OpenRouter gives in US dollars per token. Each value was multiplied by 1,000,000 and printed at full precision. Ties on input are broken by output price, then by model ID.¹

The 13 labs are OpenAI, Mistral, DeepSeek, Meta, Alibaba Qwen, Amazon, Cohere, Google, Zhipu, MiniMax, Moonshot AI, Anthropic and xAI. Free listings are excluded, which leaves 318 paid models.¹

Model IDs ending in :batch are OpenRouter’s batch listings. Each one we compared with a lab’s published batch rate matched. GPT-6 Sol’s batch listing is $1.00 input and $5.00 output on both OpenRouter and OpenAI’s batch table.^1,4 Claude Haiku 4.5’s is $0.50 and $2.50 on both OpenRouter and Anthropic’s batch table.^1,5

We then checked the listed prices against six labs’ own pricing pages, captured the same day: OpenAI, Anthropic, Google, DeepSeek, Mistral and xAI.^4,5,6,3,7,2 The other seven labs are ranked on OpenRouter’s listing alone. Prices are in US dollars, and none of the captured pages states a tax treatment for tokens bought direct. How we treat company-published figures is set out in our research methodology.

Which AI API models have the lowest input price?

OpenAI’s gpt-oss-20b lists the lowest input price of the 318, at $0.018 per million tokens, followed by Mistral Nemo at $0.019 and DeepSeek V4 Flash 0731 at $0.021.¹ The first 25 rows are below, with figures exactly as OpenRouter lists them.

Rank	Model ID	Lab	Input (USD per 1M tokens)	Output (USD per 1M tokens)	Context (tokens)
1	openai/gpt-oss-20b	OpenAI	0.018	0.09	131,072
2	mistralai/mistral-nemo	Mistral	0.019	0.03	131,072
3	deepseek/deepseek-v4-flash-0731	DeepSeek	0.021	0.32	1,310,720
4	openai/gpt-oss-20b:batch	OpenAI	0.024	0.112	131,072
5	openai/gpt-5-nano:batch	OpenAI	0.025	0.20	400,000
6	meta-llama/llama-3.2-1b-instruct	Meta	0.027	0.201	60,000
7	openai/gpt-oss-120b:batch	OpenAI	0.0296	0.136	131,072
8	qwen/qwen3.7-flash	Alibaba Qwen	0.03	0.13	1,000,000
9	amazon/nova-micro-v1	Amazon	0.035	0.14	128,000
10	deepseek/deepseek-v4.1-flash	DeepSeek	0.035	0.29	1,048,576
11	cohere/command-r7b-12-2024	Cohere	0.0375	0.15	128,000
12	meta-llama/llama-3.1-8b-instruct	Meta	0.05	0.08	131,072
13	mistralai/mistral-small-24b-instruct-2501	Mistral	0.05	0.08	32,768
14	google/gemma-3-4b-it	Google	0.05	0.10	131,072
15	google/gemma-3-12b-it	Google	0.05	0.15	131,072
16	google/gemini-2.5-flash-lite:batch	Google	0.05	0.20	1,048,576
17	openai/gpt-4.1-nano:batch	OpenAI	0.05	0.20	1,047,576
18	openai/gpt-6-luna-pro:batch	OpenAI	0.05	0.25	1,050,000
19	openai/gpt-6-luna:batch	OpenAI	0.05	0.25	1,050,000
20	meta-llama/llama-3.2-3b-instruct	Meta	0.05	0.33	131,072
21	openai/gpt-5-nano	OpenAI	0.05	0.40	400,000
22	z-ai/glm-5.3-flash:batch	Zhipu	0.06	0.20	1,048,576
23	amazon/nova-lite-v1	Amazon	0.06	0.24	300,000
24	z-ai/glm-4.7-flash	Zhipu	0.0605	0.40	200,000
25	qwen/qwen3.5-flash-02-23	Alibaba Qwen	0.065	0.26	1,000,000

Source: OpenRouter public model API, captured 28 September 2026.¹

OpenAI holds 8 of the top 25 rows, more than any other lab, and six of those eight are batch listings.¹ The first three models sit $0.003 apart on input. At a billion input tokens a month, that gap is worth $3.

Output price moves far more. Mistral Nemo lists output at $0.03, and DeepSeek V4 Flash 0731, one place below it, lists output at $0.32.¹ A workload that writes more than it reads should sort this table by the output column, not the input one.

What is the cheapest model from each lab?

The cheapest listing per lab runs from $0.018 per million input tokens for OpenAI to $1.00 for xAI.¹

Lab	Cheapest listed model	Input (USD per 1M tokens)	Output (USD per 1M tokens)
OpenAI	openai/gpt-oss-20b	0.018	0.09
Mistral	mistralai/mistral-nemo	0.019	0.03
DeepSeek	deepseek/deepseek-v4-flash-0731	0.021	0.32
Meta	meta-llama/llama-3.2-1b-instruct	0.027	0.201
Alibaba Qwen	qwen/qwen3.7-flash	0.03	0.13
Amazon	amazon/nova-micro-v1	0.035	0.14
Cohere	cohere/command-r7b-12-2024	0.0375	0.15
Google	google/gemma-3-4b-it	0.05	0.10
Zhipu	z-ai/glm-5.3-flash:batch	0.06	0.20
MiniMax	minimax/minimax-01	0.20	1.10
Moonshot AI	moonshotai/kimi-k2.5	0.45	2.25
Anthropic	anthropic/claude-haiku-4.5:batch	0.50	2.50
xAI	x-ai/grok-4.3:batch	1.00	2.00

Source: OpenRouter public model API, captured 28 September 2026.¹

Three of those floors are batch listings: Zhipu’s, Anthropic’s and xAI’s. At standard rates, Zhipu’s cheapest is GLM 4.7 Flash at $0.0605 input.¹ Anthropic’s cheapest standard listing on OpenRouter is Claude Haiku 4.5 at $1.00 input and $5.00 output. xAI’s grok-build-0.1 lists at the same $1.00 and $2.00 as the batch Grok 4.3.¹

The decision changes with the size of the job. For small, high-volume calls, nine labs list a model under $0.10 input.¹ Anthropic and xAI don’t compete at that end of the market on price.

How do flagship model prices compare side by side?

OpenAI and Anthropic list the same price for their flagships, and the other four checked labs list less. GPT-6 Astra and Claude Fable 5.1 both list at $10.00 input and $50.00 output on OpenRouter.¹ Both labs’ own pages show the same figures.^4,5

“Flagship” follows the lab’s own label or page order where the captured page gives one. The basis column says which.

Lab	Flagship model	Basis	OpenRouter input	OpenRouter output	Lab page input	Lab page output
OpenAI	openai/gpt-6-astra	First row of OpenAI’s “Flagship models” table	10.00	50.00	10.00	50.00
Anthropic	anthropic/claude-fable-5.1	First row of Anthropic’s model pricing table	10.00	50.00	10	50
Google	google/gemini-3.1-pro-preview	The Pro model in Google’s Gemini 3.x text line	2.00	12.00	2.00	12.00
xAI	x-ai/grok-4.7	xAI calls it “Our flagship model”	1.60	4.80	2.00	6.00
Mistral	mistralai/mistral-medium-3-5	Mistral’s pick “For most tasks and coding”	1.50	7.50	not on captured page	not on captured page
DeepSeek	deepseek/deepseek-v4-pro	The Pro model on DeepSeek’s two-model page	0.95526	1.91052	0.66 off-peak, 1.32 peak	1.98 off-peak, 3.96 peak

All prices in US dollars per million tokens. Sources: OpenRouter¹; OpenAI⁴; Anthropic⁵; Google, rate for prompts of 200,000 tokens or fewer⁶; xAI²; Mistral⁷; DeepSeek, cache-miss input³.

Gemini 3.1 Pro Preview lists input at $2.00, one fifth of GPT-6 Astra’s $10.00.¹ Of the six flagships, DeepSeek V4 Pro lists the lowest input and output through OpenRouter, at $0.95526 and $1.91052.¹ Bought direct, it costs $0.66 per million cache-miss input tokens off-peak and $1.32 at peak.³

Which models are cheaper than OpenAI’s, tier for tier?

At each of OpenAI’s three flagship price points, at least one other lab lists a lower input price. OpenAI’s page names those three as GPT-6 Astra, GPT-6 Sol and GPT-6 Luna.⁴ The pairings below match on price band. They aren’t a capability match, which this ranking doesn’t measure.

OpenAI tier (input / output)	Same input price	Lower listed input price
GPT-6 Astra, 10.00 / 50.00	Claude Fable 5.1, 10.00 / 50.00	Claude Opus 5.5, 4.00 / 20.00; Qwen3.8 Max Prime, 4.00 / 12.00; Kimi K3, 3.00 / 15.00
GPT-6 Sol, 2.00 / 10.00	Claude Sonnet 5, 2.00 / 10.00	Grok 4.7, 1.60 / 4.80; Mistral Medium 3.5, 1.50 / 7.50; GLM 5.3, 1.40 / 4.40; DeepSeek V4 Pro, 0.95526 / 1.91052
GPT-6 Luna, 0.10 / 0.50	Gemini 2.5 Flash-Lite, 0.10 / 0.40	DeepSeek V4.1 Flash, 0.035 / 0.29; Qwen3.7 Flash, 0.03 / 0.13

All prices in US dollars per million tokens, from OpenRouter’s public model API, captured 28 September 2026.¹

By count, 36 of the 318 paid models list input below GPT-6 Luna’s $0.10.¹ 239 list it below GPT-6 Sol’s $2.00.¹ 300 list it below GPT-6 Astra’s $10.00.¹

The decision changes when output dominates the bill. Grok 4.7 lists output at $4.80, less than half of GPT-6 Sol’s $10.00. DeepSeek V4 Pro lists it at $1.91052, under a fifth.¹ A workload that writes long answers from short prompts saves more by switching than the input column suggests.

Does OpenRouter charge the same as buying direct?

For OpenAI’s GPT-6 models, Anthropic, Google and Mistral, yes: every listed price we checked matched the lab’s own page. For xAI and DeepSeek, no, and OpenAI’s GPT-5.6 Sol doesn’t match the one table that lists it.

Model	OpenRouter input / output	Lab page input / output	Match
OpenAI gpt-6-astra	10.00 / 50.00	10.00 / 50.00	Yes
OpenAI gpt-6-sol	2.00 / 10.00	2.00 / 10.00	Yes
OpenAI gpt-6-luna	0.10 / 0.50	0.10 / 0.50	Yes
OpenAI gpt-5.3-codex	1.75 / 14.00	1.75 / 14.00	Yes
OpenAI gpt-5.6-sol	2.00 / 10.00	4.00 / 20.00 (cyber models table)	No
Anthropic claude-fable-5.1	10.00 / 50.00	10 / 50	Yes
Anthropic claude-opus-5.5	4.00 / 20.00	4 / 20	Yes
Anthropic claude-sonnet-5	2.00 / 10.00	2 / 10	Yes
Anthropic claude-haiku-4.5	1.00 / 5.00	1 / 5	Yes
Google gemini-3.1-pro-preview	2.00 / 12.00	2.00 / 12.00 (200k tokens or fewer)	Yes
Google gemini-3.8-flash	0.75 / 3.75	0.75 / 3.75 through 31 December 2026	Yes, until 2027
Google gemini-2.5-flash-lite	0.10 / 0.40	0.10 / 0.40	Yes
Mistral mistral-large-2512	0.50 / 1.50	0.5 / 1.5 (Mistral Large)	Yes
xAI grok-4.7	1.60 / 4.80	2.00 / 6.00	No, OpenRouter 20% lower
DeepSeek deepseek-v4.1-flash	0.035 / 0.29	0.15 off-peak, 0.3 peak / 0.6 off-peak, 1.2 peak	No, OpenRouter lower
DeepSeek deepseek-v4-pro	0.95526 / 1.91052	0.66 off-peak, 1.32 peak / 1.98 off-peak, 3.96 peak	No
DeepSeek deepseek-v4-pro-0813	0.2523 / 4.20	0.66 off-peak, 1.32 peak / 1.98 off-peak, 3.96 peak	No

All prices in US dollars per million tokens. Sources: OpenRouter¹; OpenAI⁴; Anthropic⁵; Google⁶; Mistral⁷; xAI²; DeepSeek, cache-miss input³.

Mistral’s batch listing for Mistral Large on OpenRouter is $0.25 input and $0.75 output.¹ That is half the standard $0.50 and $1.50, which matches Mistral’s statement that batch processing cuts the price by 50%.^1,7

DeepSeek is the lab to confirm before you buy. Its own page bills DeepSeek-V4.1-Flash at $0.15 per million cache-miss input tokens off-peak and $0.3 at peak.³ OpenRouter lists it at $0.035.¹ DeepSeek’s page also says the legacy names deepseek-v4-flash and deepseek-v4-flash-vision-exp are served by V4.1-Flash and billed at the Flash price.³ OpenRouter still lists them as separate models, at $0.14 and $0.44 input.¹ The captured sources don’t explain why the OpenRouter and DeepSeek figures differ. Confirm this in the current quote before routing volume either way.

Why do other published price lists disagree?

Because a lab’s price depends on how you call the model, and a list shows one number per model. Four conditions on the checked pages change the rate.

Prompt length. Google charges $2.00 input for Gemini 3.1 Pro Preview on prompts of 200,000 tokens or fewer, and $4.00 above that.⁶ OpenAI lists GPT-6 Sol at $2.00 for short context and $4.00 for long context.⁴ Anthropic bills Claude 4.6 and later models at standard pricing across the full 1M-token window.⁵

Processing mode. OpenAI’s batch rate for GPT-6 Sol is $1.00 input.⁴ Anthropic’s Batch API takes 50% off both input and output tokens.⁵ Mistral says batch processing cuts the price by 50% and cached input tokens cut input cost by up to 90%.⁷

Time of day and region. DeepSeek’s off-peak rates are half its peak rates, with peak hours of 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday.³ OpenAI charges a 10% uplift on data-residency endpoints for eligible models released on or after 5 March 2026.⁴ Anthropic applies a 1.1x multiplier when Claude 4.6 and later models are pinned to US-only inference.⁵

Promotions with end dates. Gemini 3.8 Flash lists at $0.75 input through 31 December 2026 and $1.50 from 1 January 2027.⁶ OpenAI says GPT-5.6 Sol’s promotional pricing runs at least through 21 November 2026.⁴ Its cyber-models table lists gpt-5.6-sol at $4.00 input and $20.00 output.⁴ OpenRouter lists openai/gpt-5.6-sol at $2.00 and $10.00.¹

What the ranking cannot tell you

This ranking measures listed price, not quality, speed or reliability. A model that costs a tenth as much can still cost more per finished task if it needs retries or longer outputs. Use one real workflow to test the claim before you move production traffic.

Price per token also assumes a token means the same thing across labs. It doesn’t. Anthropic says Claude 4.7 and later models use a newer tokenizer that produces approximately 30% more tokens for the same text.⁵ Two models with the same listed price can bill different amounts for the same prompt.

Listed prices leave out charges that sit beside the token rate. OpenAI bills web search at $10.00 per 1,000 calls plus search content tokens.⁴ Google charges $14 per 1,000 requests for grounding with Google Search, after 5,000 free a month shared across Gemini 3.x models.⁶ DeepSeek caps concurrency at 2,500 for its Flash model and 500 for V4 Pro.³ Our explainer on what controls AI inference cost covers the cost drivers behind list prices like these.

Seven of the 13 labs weren’t cross-checked against their own pages: Meta, Alibaba Qwen, Amazon, Cohere, Zhipu, MiniMax and Moonshot AI. Their rows are OpenRouter listings only. The prices are also a single day’s snapshot, and two of them carry published end dates. Other published Glamdring work sits in the research archive.

Frequently asked questions

Is there a free alternative to the OpenAI API?

Yes, with limits. Google’s Gemini API has a free tier with free input and output tokens on a limited set of models, and Google says free-tier content is used to improve its products.⁶ Gemma 4 is free of charge on that tier and has no paid rate.⁶ Mistral’s Free plan includes $10 a month in API credits.⁷ This ranking covers paid listings only, so free tiers don’t appear in the tables.

Is the OpenAI API cheaper than Anthropic’s?

At the top and middle tiers they list the same price. GPT-6 Astra and Claude Fable 5.1 both list at $10.00 input and $50.00 output.¹ GPT-6 Sol and Claude Sonnet 5 both list at $2.00 and $10.00.¹ At the low end OpenAI is far cheaper. Its cheapest model lists at $0.018 input, against $1.00 for Claude Haiku 4.5, Anthropic’s cheapest standard listing on OpenRouter.¹ Which one suits your work is a quality question this ranking doesn’t measure.

Is the OpenAI API being deprecated?

No such notice appears on OpenAI’s pricing page. It says OpenAI is winding down its fine-tuning platform, which new users can no longer access, and that fine-tuned models stay available for inference until their base models are deprecated.⁴

Will these prices change?

Some already have dates. Gemini 3.8 Flash rises to $1.50 input and $7.50 output on 1 January 2027.⁶ OpenAI’s GPT-5.6 Sol promotional pricing runs at least through 21 November 2026.⁴ DeepSeek’s page says its prices may vary and that DeepSeek reserves the right to adjust them.³ Check the lab’s page on the day you commit.