AI Infrastructure TCO Calculator

By Julien Simon | AI Operating Partner, Fortino Capital | October 18, 2026 Pricing, refreshed automatically every Sunday

Compare API costs, self-hosted GPU, and local/edge deployment for AI inference workloads.
Fill in your parameters below, then explore the analysis tabs.

Router fees
Router	Fee	Fee basis	BYOK	Notes
OpenRouter	5.5%	credit purchase (Gateway)	free up to $25k/month list-price usage, then 5%	No markup on token prices. Card credits 5.5% ($0.80 minimum), crypto 5%

Router

Fee

Fee basis

BYOK

Notes

OpenRouter

5.5%

credit purchase (Gateway)

free up to $25k/month list-price usage, then 5%

No markup on token prices. Card credits 5.5% ($0.80 minimum), crypto 5%

Router	Fee	Fee basis	BYOK	Notes
OpenRouter	5.5%	credit purchase (card)	free up to $25k/month list-price usage, then 5%	No markup on token prices. Card credits 5.5% ($0.80 minimum), crypto 5%
Opper	3%	credit purchase (Gateway)	free on Gateway	No markup on token prices. Control Plane features raise the fee to 5.5%
Requesty	5%	markup on usage	included on pay-as-you-go	Provider price + 5%. Catalog prices exclude the markup

Router

Fee

Fee basis

BYOK

Notes

OpenRouter

5.5%

credit purchase (card)

free up to $25k/month list-price usage, then 5%

No markup on token prices. Card credits 5.5% ($0.80 minimum), crypto 5%

Opper

credit purchase (Gateway)

free on Gateway

No markup on token prices. Control Plane features raise the fee to 5.5%

Requesty

markup on usage

included on pay-as-you-go

Provider price + 5%. Catalog prices exclude the markup

Model Library
Model Name	Provider	Input $/M tok	Output $/M tok	OpenRouter	Opper	Requesty	Notes
Muse Spark 1.3 Contributor	Anthropic	$0.1875	$0.6525	yes	yes	yes	Open-weights 235B MoE (22B active), Instruct. Via OpenRouter; bare slug delisted, now tracked on the qwen3-235b-a22b-2507 snapshot (median of 7 hosts). Model Studio first-party non-thinking $0.70/$2.80 (thinking output $8.40)

Model Name

Provider

Input $/M tok

Output $/M tok

OpenRouter

Opper

Requesty

Notes

Muse Spark 1.3 Contributor

Anthropic

$0.1875

$0.6525

yes

Open-weights 235B MoE (22B active), Instruct. Via OpenRouter; bare slug delisted, now tracked on the qwen3-235b-a22b-2507 snapshot (median of 7 hosts). Model Studio first-party non-thinking $0.70/$2.80 (thinking output $8.40)

Model Name	Provider	Input $/M tok	Output $/M tok	OpenRouter	Opper	Requesty	Notes
GPT-6.1 Sol	OpenAI	$2	$10	yes	yes	yes	Near-Astra performance at Sol pricing, Sep 29, 2026, 1.05M context. Long-context (>272K) $4/$15. Cached input $0.10
GPT-6 Astra	OpenAI	$10	$50	yes	yes	yes	GPT-6 flagship, 1.05M context, Sep 2026. Long-context (>272K) $20/$75. Cached input $1
GPT-6 Sol	OpenAI	$2	$10	yes	yes	yes	Balanced GPT-6 tier, 1.05M context, Sep 2026. Long-context (>272K) $4/$15. Cached input $0.20
GPT-6 Luna	OpenAI	$0.1	$0.5	yes	yes	yes	Cost GPT-6 tier, 1.05M context, Sep 2026. Long-context $0.20/$0.75
GPT-5.6 Sol	OpenAI	$4	$20	yes	yes	yes	Frontier tier, 1.05M context. Long-context (>272K) $8/$30. Promo rate at least through Nov 21, 2026
GPT-5.6 Terra	OpenAI	$2	$12	yes	yes	yes	Balanced tier, Jul 2026. Long-context $4/$18
GPT-5.6 Luna	OpenAI	$0.2	$1.2	yes	yes	yes	Cost tier, Jul 2026. Long-context $0.40/$1.80
GPT-5.6 Cyber	OpenAI	$12.5	$75				Daybreak cyber model (daybreak-red-latest), Aug 2026
GPT-5.5 Pro	OpenAI	$30	$180	yes	yes	yes	Top-tier reasoning model, Jun 2026
GPT-5.5	OpenAI	$5	$30	yes	yes	yes	Flagship, Jun 2026
GPT-5.5 Cyber	OpenAI	$12.5	$75				Cyber model variant, same rate as GPT-5.6 Cyber. Cached input $1.25
GPT-5.4 Pro	OpenAI	$30	$180	yes	yes	yes	Reasoning model, Mar 2026
GPT-5.4	OpenAI	$2.5	$15	yes	yes	yes	Previous flagship, Feb 2026
GPT-5.4 Mini	OpenAI	$0.75	$4.5	yes	yes	yes	Mid-tier 5.4 variant, Mar 2026
GPT-5.4 Nano	OpenAI	$0.2	$1.25	yes	yes	yes	Efficient 5.4 variant, Mar 2026
GPT-5.3 Codex	OpenAI	$1.75	$14	yes	yes	yes	Base gpt-5.3 retired; gpt-5.3-codex/-chat-latest remain
GPT-5.2 Pro	OpenAI	$21	$168	yes	yes		Reasoning variant, 400K context (Mar 2026 halving reverted)
GPT-5.2	OpenAI	$1.75	$14	yes	yes	yes	Dec 2025 flagship (Mar 2026 halving reverted)
GPT-5.1	OpenAI	$1.25	$10	yes	yes	yes	Coding-optimized (Mar 2026 halving reverted)
GPT-5 Pro	OpenAI	$15	$120	yes	yes	yes	Highest-cost GPT-5 reasoning tier
GPT-5	OpenAI	$1.25	$10	yes	yes	yes	Aug 2025 flagship, 400K context. Retires Dec 11, 2026
GPT-5 Mini	OpenAI	$0.25	$2	yes	yes	yes	Efficient mid-tier. Retires Dec 11, 2026
GPT-5 Nano	OpenAI	$0.05	$0.4	yes	yes	yes	Cheapest OpenAI option
GPT-4.1	OpenAI	$2	$8	yes	yes	yes	Strong all-rounder, 1M context
GPT-4.1 Mini	OpenAI	$0.4	$1.6	yes	yes	yes	Efficient mid-tier, 1M context
GPT-4.1 Nano	OpenAI	$0.1	$0.4	yes	yes	yes	Fastest & cheapest GPT-4.1
o3	OpenAI	$2	$8	yes	yes	yes	Reasoning model. Retires Dec 11, 2026
o3-mini	OpenAI	$1.1	$4.4	yes	yes	yes	Affordable reasoning. Retires Oct 23, 2026
o4-mini	OpenAI	$1.1	$4.4	yes	yes	yes	Affordable reasoning. Retires Oct 23, 2026
o1	OpenAI	$15	$60	yes	yes	yes	Legacy reasoning. Retires Oct 23, 2026
GPT-Rosalind Research	OpenAI	$5	$25				Life-sciences reasoning model (biology/drug discovery). Trusted-access program only (qualified research orgs); billing begins Oct 5, 2026. Cached input $0.50; no cache-write pricing
Claude Fable 5.1	Anthropic	$10	$50	yes	yes	yes	Most capable GA Anthropic model, 1M context, Sep 1, 2026. Cache read $0.25 (2.5%)
Claude Mythos 5.1	Anthropic	$10	$50				Limited availability (Project Glasswing); same rate as Fable 5.1
Claude Fable 5	Anthropic	$10	$50	yes	yes	yes	Previous Fable generation, 1M context, Jun 2026
Claude Mythos 5	Anthropic	$10	$50				Limited availability; same rate as Fable 5
Claude Opus 5.5	Anthropic	$4	$20	yes	yes	yes	Recommended default for most workloads, 1M context, Sep 2026. Cache read $0.20 (5%)
Claude Opus 5	Anthropic	$5	$25	yes	yes	yes	Previous Opus generation, 1M context, Jul 2026
Claude Opus 4.8	Anthropic	$5	$25	yes	yes	yes	Previous agentic coding flagship, 1M context
Claude Sonnet 5.5	Anthropic	$2	$10	yes	yes	yes	New Sonnet generation, Oct 2026; same $2/$10 rate as Sonnet 5
Claude Sonnet 5	Anthropic	$2	$10	yes	yes	yes	Intro $2/$10 now standard; Sep 2026 rise to $3/$15 cancelled
Claude Haiku 5.5	Anthropic	$0.1	$0.5	yes	yes	yes	New Haiku generation, Oct 2026, ≤100K-token prompts. Over 100K tokens: $0.50/$2.50. Cache read $0.01 (10%)
Claude Haiku 4.5	Anthropic	$1	$5	yes	yes	yes	Fast & efficient, great for routing
Claude Opus 4.7	Anthropic	$5	$25	yes	yes	yes	1M context; new tokenizer (~30% more tokens)
Claude Opus 4.6	Anthropic	$5	$25	yes	yes	yes	Previous Opus generation, 1M context
Claude Sonnet 4.6	Anthropic	$3	$15	yes	yes	yes	Previous Sonnet, 1M context
Claude Opus 4.5	Anthropic	$5	$25	yes	yes	yes	Legacy Opus
Claude Sonnet 4.5	Anthropic	$3	$15	yes	yes	yes	Deprecated Sep 30, 2026; retires Nov 30, 2026 (replacement: Sonnet 5.5)
Gemini 4 Argon	Google	$2	$10	yes	yes	yes	New Google flagship reasoning model, Oct 13, 2026. Not on Free tier. Promo rate through Nov 30, 2026; list $4/$20 from Dec 1, 2026. Context caching $0.10 (promo)/$0.20 (list)
Gemini 3.8 Flash	Google	$0.75	$3.75	yes	yes	yes	Latest Google flagship, GA Sep 2, 2026, 1M context. Promo rate; list $1.50/$7.50 from Jan 1, 2027
Gemini 3.6 Flash	Google	$0.75	$3.75	yes	yes	yes	Jul 2026. Promo rate; list $1.50/$7.50 from Jan 1, 2027
Gemini 3.5 Flash-Lite	Google	$0.3	$2.5	yes	yes	yes	Cost-efficient 3.5 variant, Jul 2026
Gemini 3.1 Pro	Google	$2	$12	yes	yes	yes	Preview; ≤200K-token prompt pricing
Gemini 3 Flash	Google	$0.5	$3	yes	yes	yes	Preview; Pro-grade reasoning at Flash speed
Gemini 3.1 Flash Lite	Google	$0.25	$1.5	yes	yes	yes	Cost-efficient 3.1 variant, Mar 2026
Gemini 2.5 Pro	Google	$1.25	$10	yes	yes	yes	Production-ready, 1M context
Gemini 2.5 Flash	Google	$0.3	$2.5	yes	yes	yes	Capable budget option
Gemini 2.5 Flash-Lite	Google	$0.1	$0.4	yes	yes	yes	Cheapest Google model
Grok 4.7	xAI	$2	$6	yes	yes	yes	New flagship, Sep 2026, 500K context; doubles above 200K-token prompts. Cached input $0.50
Grok 4.6	xAI	$2	$6	yes	yes	yes	Previous flagship, Aug 2026, 500K context; doubles above 200K-token prompts
Grok 4.5	xAI	$2	$6	yes	yes	yes	Previous flagship, Jun 2026; doubles above 200K-token prompts
Grok 4.3	xAI	$1.25	$2.5	yes	yes	yes	Mid-tier Grok
Grok 4.20	xAI	$1.25	$2.5	yes	yes		2M context (price cut from $2/$6)
Grok Build 0.1	xAI	$1	$2	yes	yes	yes	Agentic build model; doubles above 200K-token prompts
Mistral Large 4	Mistral	$0.68	$2.09		yes	yes	Open-weight multimodal flagship, v26.10, Oct 2026. Sale price; list $1.36/$4.18
Mistral Large 3	Mistral	$0.5	$1.5	yes	yes	yes	675B params, via Mistral API
Mistral Medium 3.5	Mistral	$1.5	$7.5	yes	yes	yes	Mid-tier flagship, 262K context, Jun 2026
Mistral Small 4	Mistral	$0.15	$0.6	yes	yes	yes	Hybrid reasoning, multimodal, 262K context
Codestral	Mistral	$0.3	$0.9	yes	yes	yes	Coding specialist, 256K context
Ministral 3 14B	Mistral	$0.2	$0.2				Edge tier, 262K context; flat input/output rate
Ministral 3 8B	Mistral	$0.15	$0.15				Edge tier, 262K context; flat input/output rate
Ministral 3 3B	Mistral	$0.1	$0.1	yes	yes		Smallest Ministral, 131K context; flat input/output rate
Voxtral Small	Mistral	$0.1	$0.4	yes	yes		Audio-input instruct model; text-token rates (audio priced per minute)
DeepSeek V4.1 Flash	DeepSeek	$0.15	$0.6	yes	yes	yes	First-party API (deepseek-flash), off-peak cache-miss, 1M context, Sep 10, 2026; updated to V4.1-Flash-1017 Oct 17, 2026 (no price change). Peak (01-04 & 06-10 UTC, Mon-Fri) 2x: $0.30/$1.20
DeepSeek V4 Pro	DeepSeek	$0.462	$1.386	yes	yes	yes	Phased out by DeepSeek Sep 14, 2026 (first-party requests route to V4.1 Flash). V4-Pro-0813 via OpenRouter third-party hosts
DeepSeek V4 Flash	DeepSeek	$0.03	$0.32	yes	yes	yes	Retired by DeepSeek Sep 10, 2026 (routes to V4.1 Flash). V4-Flash-0731 via OpenRouter third-party hosts
Qwen3.8 Max	Alibaba	$2	$6	yes	yes	yes	2.4T A95B multimodal, 1M context. Model Studio (Singapore) ≤1M tier
Qwen3.8 2.4T A95B	Alibaba	$2	$6	yes	yes	yes	Open-weights 2.4T MoE (95B active), 1M context. Via OpenRouter.
Qwen3.8 27B	Alibaba	$0.42	$3.0	yes	yes	yes	Open-weights 27B, Aug 2026. Via OpenRouter; Model Studio first-party $0.50/$3.00
Qwen3.8 Flash	Alibaba	$0.15	$0.47	yes	yes	yes	Model Studio (Singapore) ≤1M tier, Aug 26, 2026
Qwen3.7 Max	Alibaba	$2.5	$7.5	yes	yes	yes	Model Studio (Singapore); 50% promo ended Sep 2026
Qwen3.7 Plus	Alibaba	$0.32	$1.28	yes	yes	yes	Model Studio ≤256K non-thinking; list $0.40/$1.60 less 20%
Qwen3.7 Flash	Alibaba	$0.03	$0.13	yes			Model Studio ≤32K tier, 1M context; $0.10/$0.40 to 256K, $0.20/$0.80 above
Qwen3.6 Flash	Alibaba	$0.25	$1.5	yes			Model Studio ≤256K tier; $1/$4 above 256K
Qwen3.5 Plus	Alibaba	$0.4	$2.4	yes			Model Studio ≤256K tier; $0.50/$3 above 256K
Qwen3.5 Flash	Alibaba	$0.1	$0.4	yes			Model Studio (Singapore) budget tier
Qwen3.5 397B A17B	Alibaba	$0.55	$3.5	yes	yes	yes	Open-weights 397B MoE (17B active), vision-language. Via OpenRouter; Model Studio first-party $0.60/$3.60
Qwen3 Max	Alibaba	$1.2	$6			yes	Model Studio ≤32K tier; rises to $3/$15 at 256K
Qwen3 235B A22B	Alibaba	$0.15	$0.75	yes		yes	Open-weights 235B MoE (22B active), Instruct. Via OpenRouter; bare slug delisted, now tracked on the qwen3-235b-a22b-2507 snapshot (median of 7 hosts). Model Studio first-party non-thinking $0.70/$2.80 (thinking output $8.40)
Kimi K3	Moonshot	$3	$15	yes	yes	yes	2.8T MoE multimodal reasoning, 1M context. Cache-hit input $0.30
Kimi K2.7 Code	Moonshot	$0.95	$4	yes	yes	yes	Coding-specialised. Cache-hit input $0.19
Kimi K2.7 Code Highspeed	Moonshot	$1.9	$8			yes	Faster-serving K2.7 Code tier, 2x rate. Cache-hit input $0.38
Kimi K2.6	Moonshot	$0.95	$4	yes	yes	yes	262K context. Cache-hit input $0.16
Kimi K2.5	Moonshot	$0.532	$3.325	yes	yes	yes	Retired from Moonshot API Aug 31, 2026; third-party hosts via OpenRouter (median of 5 hosts, up from $0.45/$2.25).
MiniMax M3	MiniMax	$0.3	$1.2	yes	yes	yes	Latest MiniMax flagship, Jun 2026. Via OpenRouter.
MiniMax M2.7	MiniMax	$0.21	$0.84	yes	yes	yes	Previous flagship. Via OpenRouter.
MiniMax M2.5	MiniMax	$0.27	$1.08	yes	yes	yes	Older flagship. Via OpenRouter.
MiniMax M2-Her	MiniMax	$0.3	$1.2	yes			65K context. Via OpenRouter.
GLM 5.3	Zhipu	$1.4	$4.4	yes	yes	yes	Latest Z.ai flagship, Aug 2026. First-party API
GLM 5.3 Flash	Zhipu	$0.15	$0.5	yes	yes	yes	1M context, Aug 2026. Launch promo rate became the standard price after the Sep 9, 2026 promo window; no reversion followed
GLM 5.3 FlashX	Zhipu	$0.37	$1.25	yes		yes	High-speed serving of GLM 5.3 Flash, 1M context, Sep 2026. First-party API
GLM 5.2	Zhipu	$1.4	$4.4	yes	yes	yes	1M context, agentic/coding. Also on Mistral La Plateforme at same price
Muse Spark 1.3	Meta	$1.25	$4.25	yes	yes	yes	Meta frontier tier, 1M context, Sep 2, 2026. Cached input $0.15. Via OpenRouter.
Muse Spark 1.3 Contributor	Meta	$0.1	$0.2	yes	yes		Low-cost Muse Spark 1.3 tier, Sep 2026. Via OpenRouter.
Muse Spark 1.2	Meta	$1.25	$4.25	yes	yes		Previous Meta frontier tier, Aug 2026. Via OpenRouter.
Muse Spark 1.2 Contributor	Meta	$0.1	$0.2	yes			Low-cost Muse Spark 1.2 tier, Aug 2026. Via OpenRouter.
Muse Spark 1.1	Meta	$1.25	$4.25	yes			Meta frontier tier, 1M context, Jul 2026. Cached input $0.15. Via OpenRouter.
Muse Glimmer 30B	Meta	$0.35	$1.5	yes	yes	yes	Efficient Muse tier, Aug 2026. Via OpenRouter (raised from $0.30/$1.20).
Llama 4 Maverick	Meta	$0.1875	$0.6525	yes	yes	yes	Open-weights 400B MoE (17B active). Via OpenRouter.
Llama 4 Scout	Meta	$0.1	$0.3	yes	yes		Open-weights, efficient Llama 4 variant. Via OpenRouter.

Model Name

Provider

Input $/M tok

Output $/M tok

OpenRouter

Opper

Requesty

Notes

GPT-6.1 Sol

OpenAI

$10

yes

Near-Astra performance at Sol pricing, Sep 29, 2026, 1.05M context. Long-context (>272K) $4/$15. Cached input $0.10

GPT-6 Astra

OpenAI

$10

$50

yes

GPT-6 flagship, 1.05M context, Sep 2026. Long-context (>272K) $20/$75. Cached input $1

GPT-6 Sol

OpenAI

$10

yes

Balanced GPT-6 tier, 1.05M context, Sep 2026. Long-context (>272K) $4/$15. Cached input $0.20

GPT-6 Luna

OpenAI

$0.1

$0.5

yes

Cost GPT-6 tier, 1.05M context, Sep 2026. Long-context $0.20/$0.75

GPT-5.6 Sol

OpenAI

$20

yes

Frontier tier, 1.05M context. Long-context (>272K) $8/$30. Promo rate at least through Nov 21, 2026

GPT-5.6 Terra

OpenAI

$12

yes

Balanced tier, Jul 2026. Long-context $4/$18

GPT-5.6 Luna

OpenAI

$0.2

$1.2

yes

Cost tier, Jul 2026. Long-context $0.40/$1.80

GPT-5.6 Cyber

OpenAI

$12.5

$75

Daybreak cyber model (daybreak-red-latest), Aug 2026

GPT-5.5 Pro

OpenAI

$30

$180

yes

Top-tier reasoning model, Jun 2026

GPT-5.5

OpenAI

$30

yes

Flagship, Jun 2026

GPT-5.5 Cyber

OpenAI

$12.5

$75

Cyber model variant, same rate as GPT-5.6 Cyber. Cached input $1.25

GPT-5.4 Pro

OpenAI

$30

$180

yes

Reasoning model, Mar 2026

GPT-5.4

OpenAI

$2.5

$15

yes

Previous flagship, Feb 2026

GPT-5.4 Mini

OpenAI

$0.75

$4.5

yes

Mid-tier 5.4 variant, Mar 2026

GPT-5.4 Nano

OpenAI

$0.2

$1.25

yes

Efficient 5.4 variant, Mar 2026

GPT-5.3 Codex

OpenAI

$1.75

$14

yes

Base gpt-5.3 retired; gpt-5.3-codex/-chat-latest remain

GPT-5.2 Pro

OpenAI

$21

$168

yes

Reasoning variant, 400K context (Mar 2026 halving reverted)

GPT-5.2

OpenAI

$1.75

$14

yes

Dec 2025 flagship (Mar 2026 halving reverted)

GPT-5.1

OpenAI

$1.25

$10

yes

Coding-optimized (Mar 2026 halving reverted)

GPT-5 Pro

OpenAI

$15

$120

yes

Highest-cost GPT-5 reasoning tier

GPT-5

OpenAI

$1.25

$10

yes

Aug 2025 flagship, 400K context. Retires Dec 11, 2026

GPT-5 Mini

OpenAI

$0.25

yes

Efficient mid-tier. Retires Dec 11, 2026

GPT-5 Nano

OpenAI

$0.05

$0.4

yes

Cheapest OpenAI option

GPT-4.1

OpenAI

yes

Strong all-rounder, 1M context

GPT-4.1 Mini

OpenAI

$0.4

$1.6

yes

Efficient mid-tier, 1M context

GPT-4.1 Nano

OpenAI

$0.1

$0.4

yes

Fastest & cheapest GPT-4.1

OpenAI

yes

Reasoning model. Retires Dec 11, 2026

o3-mini

OpenAI

$1.1

$4.4

yes

Affordable reasoning. Retires Oct 23, 2026

o4-mini

OpenAI

$1.1

$4.4

yes

Affordable reasoning. Retires Oct 23, 2026

OpenAI

$15

$60

yes

Legacy reasoning. Retires Oct 23, 2026

GPT-Rosalind Research

OpenAI

$25

Life-sciences reasoning model (biology/drug discovery). Trusted-access program only (qualified research orgs); billing begins Oct 5, 2026. Cached input $0.50; no cache-write pricing

Claude Fable 5.1

Anthropic

$10

$50

yes

Most capable GA Anthropic model, 1M context, Sep 1, 2026. Cache read $0.25 (2.5%)

Claude Mythos 5.1

Anthropic

$10

$50

Limited availability (Project Glasswing); same rate as Fable 5.1

Claude Fable 5

Anthropic

$10

$50

yes

Previous Fable generation, 1M context, Jun 2026

Claude Mythos 5

Anthropic

$10

$50

Limited availability; same rate as Fable 5

Claude Opus 5.5

Anthropic

$20

yes

Recommended default for most workloads, 1M context, Sep 2026. Cache read $0.20 (5%)

Claude Opus 5

Anthropic

$25

yes

Previous Opus generation, 1M context, Jul 2026

Claude Opus 4.8

Anthropic

$25

yes

Previous agentic coding flagship, 1M context

Claude Sonnet 5.5

Anthropic

$10

yes

New Sonnet generation, Oct 2026; same $2/$10 rate as Sonnet 5

Claude Sonnet 5

Anthropic

$10

yes

Intro $2/$10 now standard; Sep 2026 rise to $3/$15 cancelled

Claude Haiku 5.5

Anthropic

$0.1

$0.5

yes

New Haiku generation, Oct 2026, ≤100K-token prompts. Over 100K tokens: $0.50/$2.50. Cache read $0.01 (10%)

Claude Haiku 4.5

Anthropic

yes

Fast & efficient, great for routing

Claude Opus 4.7

Anthropic

$25

yes

1M context; new tokenizer (~30% more tokens)

Claude Opus 4.6

Anthropic

$25

yes

Previous Opus generation, 1M context

Claude Sonnet 4.6

Anthropic

$15

yes

Previous Sonnet, 1M context

Claude Opus 4.5

Anthropic

$25

yes

Legacy Opus

Claude Sonnet 4.5

Anthropic

$15

yes

Deprecated Sep 30, 2026; retires Nov 30, 2026 (replacement: Sonnet 5.5)

Gemini 4 Argon

Google

$10

yes

New Google flagship reasoning model, Oct 13, 2026. Not on Free tier. Promo rate through Nov 30, 2026; list $4/$20 from Dec 1, 2026. Context caching $0.10 (promo)/$0.20 (list)

Gemini 3.8 Flash

Google

$0.75

$3.75

yes

Latest Google flagship, GA Sep 2, 2026, 1M context. Promo rate; list $1.50/$7.50 from Jan 1, 2027

Gemini 3.6 Flash

Google

$0.75

$3.75

yes

Jul 2026. Promo rate; list $1.50/$7.50 from Jan 1, 2027

Gemini 3.5 Flash-Lite

Google

$0.3

$2.5

yes

Cost-efficient 3.5 variant, Jul 2026

Gemini 3.1 Pro

Google

$12

yes

Preview; ≤200K-token prompt pricing

Gemini 3 Flash

Google

$0.5

yes

Preview; Pro-grade reasoning at Flash speed

Gemini 3.1 Flash Lite

Google

$0.25

$1.5

yes

Cost-efficient 3.1 variant, Mar 2026

Gemini 2.5 Pro

Google

$1.25

$10

yes

Production-ready, 1M context

Gemini 2.5 Flash

Google

$0.3

$2.5

yes

Capable budget option

Gemini 2.5 Flash-Lite

Google

$0.1

$0.4

yes

Cheapest Google model

Grok 4.7

xAI

yes

New flagship, Sep 2026, 500K context; doubles above 200K-token prompts. Cached input $0.50

Grok 4.6

xAI

yes

Previous flagship, Aug 2026, 500K context; doubles above 200K-token prompts

Grok 4.5

xAI

yes

Previous flagship, Jun 2026; doubles above 200K-token prompts

Grok 4.3

xAI

$1.25

$2.5

yes

Mid-tier Grok

Grok 4.20

xAI

$1.25

$2.5

yes

2M context (price cut from $2/$6)

Grok Build 0.1

xAI

yes

Agentic build model; doubles above 200K-token prompts

Mistral Large 4

Mistral

$0.68

$2.09

yes

Open-weight multimodal flagship, v26.10, Oct 2026. Sale price; list $1.36/$4.18

Mistral Large 3

Mistral

$0.5

$1.5

yes

675B params, via Mistral API

Mistral Medium 3.5

Mistral

$1.5

$7.5

yes

Mid-tier flagship, 262K context, Jun 2026

Mistral Small 4

Mistral

$0.15

$0.6

yes

Hybrid reasoning, multimodal, 262K context

Codestral

Mistral

$0.3

$0.9

yes

Coding specialist, 256K context

Ministral 3 14B

Mistral

$0.2

Edge tier, 262K context; flat input/output rate

Ministral 3 8B

Mistral

$0.15

Edge tier, 262K context; flat input/output rate

Ministral 3 3B

Mistral

$0.1

yes

Smallest Ministral, 131K context; flat input/output rate

Voxtral Small

Mistral

$0.1

$0.4

yes

Audio-input instruct model; text-token rates (audio priced per minute)

DeepSeek V4.1 Flash

DeepSeek

$0.15

$0.6

yes

First-party API (deepseek-flash), off-peak cache-miss, 1M context, Sep 10, 2026; updated to V4.1-Flash-1017 Oct 17, 2026 (no price change). Peak (01-04 & 06-10 UTC, Mon-Fri) 2x: $0.30/$1.20

DeepSeek V4 Pro

DeepSeek

$0.462

$1.386

yes

Phased out by DeepSeek Sep 14, 2026 (first-party requests route to V4.1 Flash). V4-Pro-0813 via OpenRouter third-party hosts

DeepSeek V4 Flash

DeepSeek

$0.03

$0.32

yes

Retired by DeepSeek Sep 10, 2026 (routes to V4.1 Flash). V4-Flash-0731 via OpenRouter third-party hosts

Qwen3.8 Max

Alibaba

yes

2.4T A95B multimodal, 1M context. Model Studio (Singapore) ≤1M tier

Qwen3.8 2.4T A95B

Alibaba

yes

Open-weights 2.4T MoE (95B active), 1M context. Via OpenRouter.

Qwen3.8 27B

Alibaba

$0.42

$3.0

yes

Open-weights 27B, Aug 2026. Via OpenRouter; Model Studio first-party $0.50/$3.00

Qwen3.8 Flash

Alibaba

$0.15

$0.47

yes

Model Studio (Singapore) ≤1M tier, Aug 26, 2026

Qwen3.7 Max

Alibaba

$2.5

$7.5

yes

Model Studio (Singapore); 50% promo ended Sep 2026

Qwen3.7 Plus

Alibaba

$0.32

$1.28

yes

Model Studio ≤256K non-thinking; list $0.40/$1.60 less 20%

Qwen3.7 Flash

Alibaba

$0.03

$0.13

yes

Model Studio ≤32K tier, 1M context; $0.10/$0.40 to 256K, $0.20/$0.80 above

Qwen3.6 Flash

Alibaba

$0.25

$1.5

yes

Model Studio ≤256K tier; $1/$4 above 256K

Qwen3.5 Plus

Alibaba

$0.4

$2.4

yes

Model Studio ≤256K tier; $0.50/$3 above 256K

Qwen3.5 Flash

Alibaba

$0.1

$0.4

yes

Model Studio (Singapore) budget tier

Qwen3.5 397B A17B

Alibaba

$0.55

$3.5

yes

Open-weights 397B MoE (17B active), vision-language. Via OpenRouter; Model Studio first-party $0.60/$3.60

Qwen3 Max

Alibaba

$1.2

yes

Model Studio ≤32K tier; rises to $3/$15 at 256K

Qwen3 235B A22B

Alibaba

$0.15

$0.75

yes

Kimi K3

Moonshot

$15

yes

2.8T MoE multimodal reasoning, 1M context. Cache-hit input $0.30

Kimi K2.7 Code

Moonshot

$0.95

yes

Coding-specialised. Cache-hit input $0.19

Kimi K2.7 Code Highspeed

Moonshot

$1.9

yes

Faster-serving K2.7 Code tier, 2x rate. Cache-hit input $0.38

Kimi K2.6

Moonshot

$0.95

yes

262K context. Cache-hit input $0.16

Kimi K2.5

Moonshot

$0.532

$3.325

yes

Retired from Moonshot API Aug 31, 2026; third-party hosts via OpenRouter (median of 5 hosts, up from $0.45/$2.25).

MiniMax M3

MiniMax

$0.3

$1.2

yes

Latest MiniMax flagship, Jun 2026. Via OpenRouter.

MiniMax M2.7

MiniMax

$0.21

$0.84

yes

Previous flagship. Via OpenRouter.

MiniMax M2.5

MiniMax

$0.27

$1.08

yes

Older flagship. Via OpenRouter.

MiniMax M2-Her

MiniMax

$0.3

$1.2

yes

65K context. Via OpenRouter.

GLM 5.3

Zhipu

$1.4

$4.4

yes

Latest Z.ai flagship, Aug 2026. First-party API

GLM 5.3 Flash

Zhipu

$0.15

$0.5

yes

1M context, Aug 2026. Launch promo rate became the standard price after the Sep 9, 2026 promo window; no reversion followed

GLM 5.3 FlashX

Zhipu

$0.37

$1.25

yes

High-speed serving of GLM 5.3 Flash, 1M context, Sep 2026. First-party API

GLM 5.2

Zhipu

$1.4

$4.4

yes

1M context, agentic/coding. Also on Mistral La Plateforme at same price

Muse Spark 1.3

Meta

$1.25

$4.25

yes

Meta frontier tier, 1M context, Sep 2, 2026. Cached input $0.15. Via OpenRouter.

Muse Spark 1.3 Contributor

Meta

$0.1

$0.2

yes

Low-cost Muse Spark 1.3 tier, Sep 2026. Via OpenRouter.

Muse Spark 1.2

Meta

$1.25

$4.25

yes

Previous Meta frontier tier, Aug 2026. Via OpenRouter.

Muse Spark 1.2 Contributor

Meta

$0.1

$0.2

yes

Low-cost Muse Spark 1.2 tier, Aug 2026. Via OpenRouter.

Muse Spark 1.1

Meta

$1.25

$4.25

yes

Meta frontier tier, 1M context, Jul 2026. Cached input $0.15. Via OpenRouter.

Muse Glimmer 30B

Meta

$0.35

$1.5

yes

Efficient Muse tier, Aug 2026. Via OpenRouter (raised from $0.30/$1.20).

Llama 4 Maverick

Meta

$0.1875

$0.6525

yes

Open-weights 400B MoE (17B active). Via OpenRouter.

Llama 4 Scout

Meta

$0.1

$0.3

yes

Open-weights, efficient Llama 4 variant. Via OpenRouter.

GPU Instance Pricing (October 18, 2026)
Instance	Provider	GPU	$/hr	VRAM (GB)	Notes
GCP - RTX PRO 6000 - g4-standard-192	CoreWeave	RTX PRO 4500	14.24	192	Marketplace median, verified hosts (n=12-13, 3.06-3.11 across all 6 samples from two independent sessions; up sharply from 1.87 as n grew from 5 to 12, a deeper offer pool)

Instance

Provider

GPU

$/hr

VRAM (GB)

Notes

GCP - RTX PRO 6000 - g4-standard-192

CoreWeave

RTX PRO 4500

14.24

192

Marketplace median, verified hosts (n=12-13, 3.06-3.11 across all 6 samples from two independent sessions; up sharply from 1.87 as n grew from 5 to 12, a deeper offer pool)

Instance	Provider	GPU	$/hr	VRAM (GB)	Notes
AWS - A100 40GB - p4d.24xlarge	AWS	A100	2.74	40	8-GPU node, ~$21.96/hr total
AWS - A100 80GB - p4de.24xlarge	AWS	A100	3.43	80	8-GPU node, ~$27.45/hr total
AWS - B200 - p6-b200.48xlarge	AWS	B200	14.24	192	8-GPU node, ~$113.93/hr total
AWS - B300 - p6-b300.48xlarge	AWS	B300	17.8	288	8-GPU node, ~$142.42/hr total
AWS - H100 SXM - p5.4xlarge	AWS	H100	6.88	80	1-GPU instance
AWS - H100 SXM - p5.48xlarge	AWS	H100	6.88	80	8-GPU node, ~$55.04/hr total
AWS - H200 - p5en.48xlarge	AWS	H200	7.91	141	8-GPU node, ~$63.30/hr total
AWS - L4 - g6.xlarge	AWS	L4	0.8	24	1-GPU instance
AWS - L4 - g6.2xlarge	AWS	L4	0.98	24	1-GPU instance
AWS - L4 - g6.4xlarge	AWS	L4	1.32	24	1-GPU instance
AWS - L4 - g6.8xlarge	AWS	L4	2.01	24	1-GPU instance
AWS - L4 - g6.12xlarge	AWS	L4	1.15	24	4-GPU node, ~$4.60/hr total
AWS - L4 - g6.16xlarge	AWS	L4	3.4	24	1-GPU instance
AWS - L4 - g6.24xlarge	AWS	L4	1.67	24	4-GPU node, ~$6.68/hr total
AWS - L4 - g6.48xlarge	AWS	L4	1.67	24	8-GPU node, ~$13.35/hr total
AWS - L4 - gr6.4xlarge	AWS	L4	1.54	24	1-GPU instance, RAM-heavy (1:8 vCPU:RAM)
AWS - L4 - gr6.8xlarge	AWS	L4	2.45	24	1-GPU instance, RAM-heavy (1:8 vCPU:RAM)
AWS - L40S - g6e.xlarge	AWS	L40S	1.86	48	1-GPU instance
AWS - L40S - g6e.2xlarge	AWS	L40S	2.24	48	1-GPU instance
AWS - L40S - g6e.4xlarge	AWS	L40S	3	48	1-GPU instance
AWS - L40S - g6e.8xlarge	AWS	L40S	4.53	48	1-GPU instance
AWS - L40S - g6e.12xlarge	AWS	L40S	2.62	48	4-GPU node, ~$10.49/hr total
AWS - L40S - g6e.16xlarge	AWS	L40S	7.58	48	1-GPU instance
AWS - L40S - g6e.24xlarge	AWS	L40S	3.77	48	4-GPU node, ~$15.07/hr total
AWS - L40S - g6e.48xlarge	AWS	L40S	3.77	48	8-GPU node, ~$30.13/hr total
AWS - RTX PRO 4500 - g7.2xlarge	AWS	RTX PRO 4500	2.52	32	1-GPU instance
AWS - RTX PRO 4500 - g7.4xlarge	AWS	RTX PRO 4500	3.04	32	1-GPU instance
AWS - RTX PRO 4500 - g7.8xlarge	AWS	RTX PRO 4500	4.09	32	1-GPU instance
AWS - RTX PRO 4500 - g7.12xlarge	AWS	RTX PRO 4500	3.56	32	2-GPU node, ~$7.13/hr total
AWS - RTX PRO 4500 - g7.24xlarge	AWS	RTX PRO 4500	3.56	32	4-GPU node, ~$14.26/hr total
AWS - RTX PRO 4500 - g7.48xlarge	AWS	RTX PRO 4500	3.56	32	8-GPU node, ~$28.51/hr total
AWS - RTX PRO 6000 - g7e.2xlarge	AWS	RTX PRO 6000	3.36	96	1-GPU instance
AWS - RTX PRO 6000 - g7e.4xlarge	AWS	RTX PRO 6000	4	96	1-GPU instance
AWS - RTX PRO 6000 - g7e.8xlarge	AWS	RTX PRO 6000	5.27	96	1-GPU instance
AWS - RTX PRO 6000 - g7e.12xlarge	AWS	RTX PRO 6000	4.14	96	2-GPU node, ~$8.29/hr total
AWS - RTX PRO 6000 - g7e.24xlarge	AWS	RTX PRO 6000	4.14	96	4-GPU node, ~$16.57/hr total
AWS - RTX PRO 6000 - g7e.48xlarge	AWS	RTX PRO 6000	4.14	96	8-GPU node, ~$33.14/hr total
Azure - A100 40GB - ND96asr v4	Azure	A100	3.4	40	8-GPU node, ~$27.20/hr total
Azure - A100 80GB - NC24ads A100 v4	Azure	A100	3.67	80	1-GPU instance
Azure - A100 80GB - NC48ads A100 v4	Azure	A100	3.67	80	2-GPU node, ~$7.35/hr total
Azure - A100 80GB - NC96ads A100 v4	Azure	A100	3.67	80	4-GPU node, ~$14.69/hr total
Azure - A100 80GB - ND96amsr v4	Azure	A100	4.1	80	8-GPU node, ~$32.77/hr total (NDm A100 v4)
Azure - GB200 - ND GB200 v6	Azure	GB200	27.04	186	4-GPU node, ~$108.16/hr total
Azure - H100 NVL - NC40ads H100 v5	Azure	H100	6.98	94	1-GPU instance
Azure - H100 NVL - NC80adis H100 v5	Azure	H100	6.98	94	2-GPU node, ~$13.96/hr total
Azure - H100 - ND96is H100 v5	Azure	H100	11.06	80	8-GPU node, ~$88.49/hr total; no-InfiniBand variant (is/is_flex/noIB same price)
Azure - H100 - ND96isr H100 v5	Azure	H100	12.29	80	8-GPU node, ~$98.32/hr total
Azure - H200 - ND H200 v5	Azure	H200	10.6	141	8-GPU node, ~$84.80/hr total (East US 2)
Azure - MI300X - ND MI300X v5	Azure	MI300X	6	192	8-GPU node, ~$48.00/hr total (East US 2)
Azure - RTX PRO 6000 - NC144ds v6	Azure	RTX PRO 6000	6.38	96	1-GPU size; NC144lds (local-disk) $5.50
Azure - RTX PRO 6000 - NC288ds v6	Azure	RTX PRO 6000	5.5	96	2-GPU node, ~$11.00/hr total
CoreWeave - A100 80GB	CoreWeave	A100	2.7	80	8-GPU node, ~$21.60/hr total
CoreWeave - B200	CoreWeave	B200	8.6	192	8-GPU node, ~$68.80/hr total
CoreWeave - GB200 NVL72	CoreWeave	GB200	10.5	186	4-GPU node, ~$42.00/hr total
CoreWeave - GH200	CoreWeave	GH200	6.5	96	1-GPU instance
CoreWeave - H100 SXM	CoreWeave	H100	6.16	80	8-GPU node, ~$49.24/hr total
CoreWeave - H200	CoreWeave	H200	6.31	141	8-GPU node, ~$50.44/hr total
CoreWeave - L40	CoreWeave	L40	1.25	48	8-GPU node, ~$10.00/hr total
CoreWeave - L40S	CoreWeave	L40S	2.25	48	8-GPU node, ~$18.00/hr total
CoreWeave - RTX PRO 6000	CoreWeave	RTX PRO 6000	2.5	96	8-GPU node, ~$20.00/hr total
Crusoe - A100 SXM 80GB	Crusoe	A100	2.3	80	On-demand
Crusoe - A100 PCIe 80GB	Crusoe	A100	2	80	On-demand
Crusoe - B200	Crusoe	B200	8.5	192	On-demand; new self-serve row as of the Oct 18, 2026 refresh. Crusoe's page lists it as "180GB HGX" but other providers record B200 at 192GB
Crusoe - B300	Crusoe	B300	9.9	288	HGX B300, on-demand; new self-serve row as of the Oct 18, 2026 refresh
Crusoe - H100 HGX	Crusoe	H100	4.5	80	On-demand (up from 3.90)
Crusoe - H200 HGX	Crusoe	H200	5.4	141	On-demand (up from 4.29)
Crusoe - L40S	Crusoe	L40S	1.5	48	On-demand
Crusoe - MI300X	Crusoe	MI300X	3.45	192	On-demand
GCP - A100 40GB - a2-highgpu-1g	GCP	A100	3.67	40	1-GPU instance
GCP - A100 40GB - a2-highgpu-2g	GCP	A100	3.67	40	2-GPU node, ~$7.35/hr total
GCP - A100 40GB - a2-highgpu-4g	GCP	A100	3.67	40	4-GPU node, ~$14.69/hr total
GCP - A100 40GB - a2-highgpu-8g	GCP	A100	3.67	40	8-GPU node, ~$29.39/hr total
GCP - A100 40GB - a2-megagpu-16g	GCP	A100	3.48	40	16-GPU node, ~$55.74/hr total
GCP - A100 80GB - a2-ultragpu-1g	GCP	A100	5.07	80	1-GPU instance
GCP - A100 80GB - a2-ultragpu-2g	GCP	A100	5.07	80	2-GPU node, ~$10.14/hr total
GCP - A100 80GB - a2-ultragpu-4g	GCP	A100	5.07	80	4-GPU node, ~$20.28/hr total
GCP - A100 80GB - a2-ultragpu-8g	GCP	A100	5.07	80	8-GPU node, ~$40.55/hr total
GCP - H100 SXM - a3-highgpu-8g	GCP	H100	11.06	80	8-GPU node, ~$88.49/hr total
GCP - H100 SXM - a3-megagpu-8g	GCP	H100	11.68	80	8-GPU node, ~$93.40/hr total (Mega, GPUDirect-TCPXO)
GCP - H200 - a3-ultragpu-8g	GCP	H200	10.6	141	8-GPU node, ~$84.81/hr total
GCP - L4 - g2-standard-4	GCP	L4	0.71	24	1-GPU instance
GCP - L4 - g2-standard-8	GCP	L4	0.85	24	1-GPU instance
GCP - L4 - g2-standard-12	GCP	L4	1	24	1-GPU instance
GCP - L4 - g2-standard-16	GCP	L4	1.15	24	1-GPU instance
GCP - L4 - g2-standard-24	GCP	L4	1	24	2-GPU node, ~$2.00/hr total
GCP - L4 - g2-standard-32	GCP	L4	1.73	24	1-GPU instance
GCP - L4 - g2-standard-48	GCP	L4	1	24	4-GPU node, ~$4.00/hr total
GCP - L4 - g2-standard-96	GCP	L4	1	24	8-GPU node, ~$8.00/hr total
GCP - RTX PRO 6000 - g4-standard-48	GCP	RTX PRO 6000	4.5	96	1-GPU instance
GCP - RTX PRO 6000 - g4-standard-96	GCP	RTX PRO 6000	4.5	96	2-GPU node, ~$9.00/hr total
GCP - RTX PRO 6000 - g4-standard-192	GCP	RTX PRO 6000	4.5	96	4-GPU node, ~$18.00/hr total
GCP - RTX PRO 6000 - g4-standard-384	GCP	RTX PRO 6000	4.5	96	8-GPU node, ~$36.00/hr total
Lambda - A100 SXM 40GB	Lambda	A100	1.99	40	1-GPU instance
Lambda - A100 SXM 80GB	Lambda	A100	2.79	80	8-GPU node pricing
Lambda - B200 SXM	Lambda	B200	6.69	192	8-GPU node pricing
Lambda - GH200	Lambda	GH200	2.29	96	1-GPU instance
Lambda - H100 PCIe	Lambda	H100	3.29	80	1-GPU instance
Lambda - H100 SXM	Lambda	H100	3.99	80	8-GPU node pricing
RunPod - A100 PCIe 80GB	RunPod	A100	1.79	80	Secure Cloud on-demand (up from 1.59)
RunPod - A100 SXM 80GB	RunPod	A100	1.79	80	Secure Cloud on-demand (up from 1.59)
RunPod - B200	RunPod	B200	7.99	192	Secure Cloud on-demand (up from 6.79)
RunPod - B300	RunPod	B300	8.99	288	Secure Cloud on-demand (up from 7.89)
RunPod - H100 NVL	RunPod	H100	3.19	94	Secure Cloud on-demand
RunPod - H100 PCIe	RunPod	H100	2.89	80	Secure Cloud on-demand
RunPod - H100 SXM	RunPod	H100	3.99	80	Secure Cloud on-demand (up from 3.49)
RunPod - H200 SXM	RunPod	H200	5.29	141	Secure Cloud on-demand (up from 4.59)
RunPod - L4	RunPod	L4	0.59	24	Secure Cloud on-demand (up from 0.49)
RunPod - L40S	RunPod	L40S	1.09	48	Secure Cloud on-demand
RunPod - RTX PRO 6000	RunPod	RTX PRO 6000	2.49	96	Secure Cloud on-demand (up from 2.09)
Together - B200	Together	B200	8.99	192	Dedicated Endpoints on-demand; GPU Clusters tier is $8.19
Together - B300	Together	B300	9.99	288	GPU Clusters tier on-demand; self-serve pricing returned after being Contact-us-only at the Sep 27, 2026 refresh
Together - H100	Together	H100	3.99	80	GPU Clusters tier on-demand; Dedicated Endpoints tier is $5.49. No promo end-date language found on the page as of Oct 4, 2026 (previously noted as expiring Sep 30, 2026)
Together - H200	Together	H200	5.99	141	GPU Clusters tier on-demand; self-serve pricing returned after being Contact-us-only at the Sep 27, 2026 refresh
Vast.ai - A100 SXM 80GB	Vast.ai	A100	1.59	80	Marketplace median, verified hosts, 80GB only (n=9-12, 6 samples from two independent sessions mostly 1.59-1.60 with one 1.37 outlier; up sharply from 1.08)
Vast.ai - B200	Vast.ai	B200	9	192	Marketplace median, verified hosts (n=9, exactly 9.0026 across all 6 samples from two independent sessions; up from 7.81 as n dropped from 27
Vast.ai - H100 SXM	Vast.ai	H100	3.07	80	Marketplace median, verified hosts (n=12-13, 3.06-3.11 across all 6 samples from two independent sessions; up sharply from 1.87 as n grew from 5 to 12, a deeper offer pool)
Vast.ai - H200	Vast.ai	H200	5	141	Marketplace median, verified hosts (n=3, exactly 5.0011 across all 6 samples from two independent sessions, identical offer set each time; unchanged but thin market
Vast.ai - L4	Vast.ai	L4	0.36	24	Marketplace median, verified hosts (n=10, exactly 0.3613 across all 6 samples from two independent sessions; up from 0.33)
Vast.ai - L40S	Vast.ai	L40S	0.8	48	Marketplace median, verified hosts (n=9, stable across samples; unchanged)
Vast.ai - RTX PRO 6000	Vast.ai	RTX PRO 6000	1.59	96	Marketplace median, Server Edition (n=16-20, converged 1.5872-1.5875 across samples from two independent sessions; up from 1.47)

Instance

Provider

GPU

$/hr

VRAM (GB)

Notes

AWS - A100 40GB - p4d.24xlarge

AWS

A100

2.74

8-GPU node, ~$21.96/hr total

AWS - A100 80GB - p4de.24xlarge

AWS

A100

3.43

8-GPU node, ~$27.45/hr total

AWS - B200 - p6-b200.48xlarge

AWS

B200

14.24

192

8-GPU node, ~$113.93/hr total

AWS - B300 - p6-b300.48xlarge

AWS

B300

17.8

288

8-GPU node, ~$142.42/hr total

AWS - H100 SXM - p5.4xlarge

AWS

H100

6.88

1-GPU instance

AWS - H100 SXM - p5.48xlarge

AWS

H100

6.88

8-GPU node, ~$55.04/hr total

AWS - H200 - p5en.48xlarge

AWS

H200

7.91

141

8-GPU node, ~$63.30/hr total

AWS - L4 - g6.xlarge

AWS

0.8

1-GPU instance

AWS - L4 - g6.2xlarge

AWS

0.98

1-GPU instance

AWS - L4 - g6.4xlarge

AWS

1.32

1-GPU instance

AWS - L4 - g6.8xlarge

AWS

2.01

1-GPU instance

AWS - L4 - g6.12xlarge

AWS

1.15

4-GPU node, ~$4.60/hr total

AWS - L4 - g6.16xlarge

AWS

3.4

1-GPU instance

AWS - L4 - g6.24xlarge

AWS

1.67

4-GPU node, ~$6.68/hr total

AWS - L4 - g6.48xlarge

AWS

1.67

8-GPU node, ~$13.35/hr total

AWS - L4 - gr6.4xlarge

AWS

1.54

1-GPU instance, RAM-heavy (1:8 vCPU:RAM)

AWS - L4 - gr6.8xlarge

AWS

2.45

1-GPU instance, RAM-heavy (1:8 vCPU:RAM)

AWS - L40S - g6e.xlarge

AWS

L40S

1.86

1-GPU instance

AWS - L40S - g6e.2xlarge

AWS

L40S

2.24

1-GPU instance

AWS - L40S - g6e.4xlarge

AWS

L40S

1-GPU instance

AWS - L40S - g6e.8xlarge

AWS

L40S

4.53

1-GPU instance

AWS - L40S - g6e.12xlarge

AWS

L40S

2.62

4-GPU node, ~$10.49/hr total

AWS - L40S - g6e.16xlarge

AWS

L40S

7.58

1-GPU instance

AWS - L40S - g6e.24xlarge

AWS

L40S

3.77

4-GPU node, ~$15.07/hr total

AWS - L40S - g6e.48xlarge

AWS

L40S

3.77

8-GPU node, ~$30.13/hr total

AWS - RTX PRO 4500 - g7.2xlarge

AWS

RTX PRO 4500

2.52

1-GPU instance

AWS - RTX PRO 4500 - g7.4xlarge

AWS

RTX PRO 4500

3.04

1-GPU instance

AWS - RTX PRO 4500 - g7.8xlarge

AWS

RTX PRO 4500

4.09

1-GPU instance

AWS - RTX PRO 4500 - g7.12xlarge

AWS

RTX PRO 4500

3.56

2-GPU node, ~$7.13/hr total

AWS - RTX PRO 4500 - g7.24xlarge

AWS

RTX PRO 4500

3.56

4-GPU node, ~$14.26/hr total

AWS - RTX PRO 4500 - g7.48xlarge

AWS

RTX PRO 4500

3.56

8-GPU node, ~$28.51/hr total

AWS - RTX PRO 6000 - g7e.2xlarge

AWS

RTX PRO 6000

3.36

1-GPU instance

AWS - RTX PRO 6000 - g7e.4xlarge

AWS

RTX PRO 6000

1-GPU instance

AWS - RTX PRO 6000 - g7e.8xlarge

AWS

RTX PRO 6000

5.27

1-GPU instance

AWS - RTX PRO 6000 - g7e.12xlarge

AWS

RTX PRO 6000

4.14

2-GPU node, ~$8.29/hr total

AWS - RTX PRO 6000 - g7e.24xlarge

AWS

RTX PRO 6000

4.14

4-GPU node, ~$16.57/hr total

AWS - RTX PRO 6000 - g7e.48xlarge

AWS

RTX PRO 6000

4.14

8-GPU node, ~$33.14/hr total

Azure - A100 40GB - ND96asr v4

Azure

A100

3.4

8-GPU node, ~$27.20/hr total

Azure - A100 80GB - NC24ads A100 v4

Azure

A100

3.67

1-GPU instance

Azure - A100 80GB - NC48ads A100 v4

Azure

A100

3.67

2-GPU node, ~$7.35/hr total

Azure - A100 80GB - NC96ads A100 v4

Azure

A100

3.67

4-GPU node, ~$14.69/hr total

Azure - A100 80GB - ND96amsr v4

Azure

A100

4.1

8-GPU node, ~$32.77/hr total (NDm A100 v4)

Azure - GB200 - ND GB200 v6

Azure

GB200

27.04

186

4-GPU node, ~$108.16/hr total

Azure - H100 NVL - NC40ads H100 v5

Azure

H100

6.98

1-GPU instance

Azure - H100 NVL - NC80adis H100 v5

Azure

H100

6.98

2-GPU node, ~$13.96/hr total

Azure - H100 - ND96is H100 v5

Azure

H100

11.06

8-GPU node, ~$88.49/hr total; no-InfiniBand variant (is/is_flex/noIB same price)

Azure - H100 - ND96isr H100 v5

Azure

H100

12.29

8-GPU node, ~$98.32/hr total

Azure - H200 - ND H200 v5

Azure

H200

10.6

141

8-GPU node, ~$84.80/hr total (East US 2)

Azure - MI300X - ND MI300X v5

Azure

MI300X

192

8-GPU node, ~$48.00/hr total (East US 2)

Azure - RTX PRO 6000 - NC144ds v6

Azure

RTX PRO 6000

6.38

1-GPU size; NC144lds (local-disk) $5.50

Azure - RTX PRO 6000 - NC288ds v6

Azure

RTX PRO 6000

5.5

2-GPU node, ~$11.00/hr total

CoreWeave - A100 80GB

CoreWeave

A100

2.7

8-GPU node, ~$21.60/hr total

CoreWeave - B200

CoreWeave

B200

8.6

192

8-GPU node, ~$68.80/hr total

CoreWeave - GB200 NVL72

CoreWeave

GB200

10.5

186

4-GPU node, ~$42.00/hr total

CoreWeave - GH200

CoreWeave

GH200

6.5

1-GPU instance

CoreWeave - H100 SXM

CoreWeave

H100

6.16

8-GPU node, ~$49.24/hr total

CoreWeave - H200

CoreWeave

H200

6.31

141

8-GPU node, ~$50.44/hr total

CoreWeave - L40

CoreWeave

L40

1.25

8-GPU node, ~$10.00/hr total

CoreWeave - L40S

CoreWeave

L40S

2.25

8-GPU node, ~$18.00/hr total

CoreWeave - RTX PRO 6000

CoreWeave

RTX PRO 6000

2.5

8-GPU node, ~$20.00/hr total

Crusoe - A100 SXM 80GB

Crusoe

A100

2.3

On-demand

Crusoe - A100 PCIe 80GB

Crusoe

A100

On-demand

Crusoe - B200

Crusoe

B200

8.5

192

On-demand; new self-serve row as of the Oct 18, 2026 refresh. Crusoe's page lists it as "180GB HGX" but other providers record B200 at 192GB

Crusoe - B300

Crusoe

B300

9.9

288

HGX B300, on-demand; new self-serve row as of the Oct 18, 2026 refresh

Crusoe - H100 HGX

Crusoe

H100

4.5

On-demand (up from 3.90)

Crusoe - H200 HGX

Crusoe

H200

5.4

141

On-demand (up from 4.29)

Crusoe - L40S

Crusoe

L40S

1.5

On-demand

Crusoe - MI300X

Crusoe

MI300X

3.45

192

On-demand

GCP - A100 40GB - a2-highgpu-1g

GCP

A100

3.67

1-GPU instance

GCP - A100 40GB - a2-highgpu-2g

GCP

A100

3.67

2-GPU node, ~$7.35/hr total

GCP - A100 40GB - a2-highgpu-4g

GCP

A100

3.67

4-GPU node, ~$14.69/hr total

GCP - A100 40GB - a2-highgpu-8g

GCP

A100

3.67

8-GPU node, ~$29.39/hr total

GCP - A100 40GB - a2-megagpu-16g

GCP

A100

3.48

16-GPU node, ~$55.74/hr total

GCP - A100 80GB - a2-ultragpu-1g

GCP

A100

5.07

1-GPU instance

GCP - A100 80GB - a2-ultragpu-2g

GCP

A100

5.07

2-GPU node, ~$10.14/hr total

GCP - A100 80GB - a2-ultragpu-4g

GCP

A100

5.07

4-GPU node, ~$20.28/hr total

GCP - A100 80GB - a2-ultragpu-8g

GCP

A100

5.07

8-GPU node, ~$40.55/hr total

GCP - H100 SXM - a3-highgpu-8g

GCP

H100

11.06

8-GPU node, ~$88.49/hr total

GCP - H100 SXM - a3-megagpu-8g

GCP

H100

11.68

8-GPU node, ~$93.40/hr total (Mega, GPUDirect-TCPXO)

GCP - H200 - a3-ultragpu-8g

GCP

H200

10.6

141

8-GPU node, ~$84.81/hr total

GCP - L4 - g2-standard-4

GCP

0.71

1-GPU instance

GCP - L4 - g2-standard-8

GCP

0.85

1-GPU instance

GCP - L4 - g2-standard-12

GCP

1-GPU instance

GCP - L4 - g2-standard-16

GCP

1.15

1-GPU instance

GCP - L4 - g2-standard-24

GCP

2-GPU node, ~$2.00/hr total

GCP - L4 - g2-standard-32

GCP

1.73

1-GPU instance

GCP - L4 - g2-standard-48

GCP

4-GPU node, ~$4.00/hr total

GCP - L4 - g2-standard-96

GCP

8-GPU node, ~$8.00/hr total

GCP - RTX PRO 6000 - g4-standard-48

GCP

RTX PRO 6000

4.5

1-GPU instance

GCP - RTX PRO 6000 - g4-standard-96

GCP

RTX PRO 6000

4.5

2-GPU node, ~$9.00/hr total

GCP - RTX PRO 6000 - g4-standard-192

GCP

RTX PRO 6000

4.5

4-GPU node, ~$18.00/hr total

GCP - RTX PRO 6000 - g4-standard-384

GCP

RTX PRO 6000

4.5

8-GPU node, ~$36.00/hr total

Lambda - A100 SXM 40GB

Lambda

A100

1.99

1-GPU instance

Lambda - A100 SXM 80GB

Lambda

A100

2.79

8-GPU node pricing

Lambda - B200 SXM

Lambda

B200

6.69

192

8-GPU node pricing

Lambda - GH200

Lambda

GH200

2.29

1-GPU instance

Lambda - H100 PCIe

Lambda

H100

3.29

1-GPU instance

Lambda - H100 SXM

Lambda

H100

3.99

8-GPU node pricing

RunPod - A100 PCIe 80GB

RunPod

A100

1.79

Secure Cloud on-demand (up from 1.59)

RunPod - A100 SXM 80GB

RunPod

A100

1.79

Secure Cloud on-demand (up from 1.59)

RunPod - B200

RunPod

B200

7.99

192

Secure Cloud on-demand (up from 6.79)

RunPod - B300

RunPod

B300

8.99

288

Secure Cloud on-demand (up from 7.89)

RunPod - H100 NVL

RunPod

H100

3.19

Secure Cloud on-demand

RunPod - H100 PCIe

RunPod

H100

2.89

Secure Cloud on-demand

RunPod - H100 SXM

RunPod

H100

3.99

Secure Cloud on-demand (up from 3.49)

RunPod - H200 SXM

RunPod

H200

5.29

141

Secure Cloud on-demand (up from 4.59)

RunPod - L4

RunPod

0.59

Secure Cloud on-demand (up from 0.49)

RunPod - L40S

RunPod

L40S

1.09

Secure Cloud on-demand

RunPod - RTX PRO 6000

RunPod

RTX PRO 6000

2.49

Secure Cloud on-demand (up from 2.09)

Together - B200

Together

B200

8.99

192

Dedicated Endpoints on-demand; GPU Clusters tier is $8.19

Together - B300

Together

B300

9.99

288

GPU Clusters tier on-demand; self-serve pricing returned after being Contact-us-only at the Sep 27, 2026 refresh

Together - H100

Together

H100

3.99

GPU Clusters tier on-demand; Dedicated Endpoints tier is $5.49. No promo end-date language found on the page as of Oct 4, 2026 (previously noted as expiring Sep 30, 2026)

Together - H200

Together

H200

5.99

141

GPU Clusters tier on-demand; self-serve pricing returned after being Contact-us-only at the Sep 27, 2026 refresh

Vast.ai - A100 SXM 80GB

Vast.ai

A100

1.59

Marketplace median, verified hosts, 80GB only (n=9-12, 6 samples from two independent sessions mostly 1.59-1.60 with one 1.37 outlier; up sharply from 1.08)

Vast.ai - B200

Vast.ai

B200

192

Marketplace median, verified hosts (n=9, exactly 9.0026 across all 6 samples from two independent sessions; up from 7.81 as n dropped from 27

Vast.ai - H100 SXM

Vast.ai

H100

3.07

Marketplace median, verified hosts (n=12-13, 3.06-3.11 across all 6 samples from two independent sessions; up sharply from 1.87 as n grew from 5 to 12, a deeper offer pool)

Vast.ai - H200

Vast.ai

H200

141

Marketplace median, verified hosts (n=3, exactly 5.0011 across all 6 samples from two independent sessions, identical offer set each time; unchanged but thin market

Vast.ai - L4

Vast.ai

0.36

Marketplace median, verified hosts (n=10, exactly 0.3613 across all 6 samples from two independent sessions; up from 0.33)

Vast.ai - L40S

Vast.ai

L40S

0.8

Marketplace median, verified hosts (n=9, stable across samples; unchanged)

Vast.ai - RTX PRO 6000

Vast.ai

RTX PRO 6000

1.59

Marketplace median, Server Edition (n=16-20, converged 1.5872-1.5875 across samples from two independent sessions; up from 1.47)

AI Infrastructure TCO Calculator

Usage Parameters

API Pricing (per 1M tokens)

Self-Hosted GPU Parameters

Local / Edge Parameters

LLM API Cost Analysis

Self-Hosted GPU Cost Analysis

Local / Edge Deployment Cost Analysis

Side-by-Side Comparison

Model Library — October 18, 2026 Pricing

Routers and gateways

GPU Instance Library — October 18, 2026 Pricing