LLM API Pricing Comparison by Model and Provider

Gemini 3.8 Flash rates are promotional through 2026-12-31 and rise to $1.50 input and $7.50 output from 2027-01-01. GPT-6.1 Sol rates apply up to 272,000 input tokens; longer prompts are billed at 2× input and 1.5× output.⁵⁴

— means no public serverless rate for that model in this comparison. Groq and Cerebras serve none of the three current models. DeepInfra is shown at list price; its current promotion is in its card.^22241213

On GLM-5.3, DeepInfra’s list input is 36% below the standard rate but its output only 9% below, so its lead shrinks from 29% to 10% when output dominates. On DeepSeek V4.1 Flash 1017 its output is 50% below standard, so the lead grows from 38% to 50%. Caching breaks the GLM-5.3-Flash tie: CoreWeave’s $0.05 cached rate puts the cached request at $0.00120, against $0.00104 elsewhere.¹²¹³³¹

1. DeepInfra

   List rates below the standard card on GLM-5.3 and DeepSeek V4.1 Flash 1017, plus a current promotion and extra service tiers.^12131418

   GLM-5.3

   $0.20

   DeepSeek V4.1 Flash 1017

   $0.006

   GLM-5.3-Flash

   $0.03

   Strengths

   Its lower $0.90 GLM-5.3 input rate outweighs a cached rate above Baseten’s, so the cached GLM-5.3 request costs $0.00740, the lowest of any provider here.^12131418

   Limitations

   On GLM-5.3 the discount sits mostly on input, so output-heavy requests save only 10%; Flex (0.8×) and, on DeepSeek V4.1 Flash 1017, Priority (1.5×) tiers are priced off the current rate.^12131418

   DeepInfra currently bills a promotional $0.5625 / $2.50 / $0.125 on GLM-5.3 and $0.098 / $0.294 / $0.003 on DeepSeek V4.1 Flash 1017; the matrix keeps list prices because promotions expire.¹²¹³

   ^12131418

2. Baseten

   Standard input and output rates, the lowest GLM-5.3 cached-input rate, and a higher-priced Fast tier.¹⁸³

   GLM-5.3

   $0.14

   DeepSeek V4.1 Flash 1017

   $0.007

   GLM-5.3-Flash

   $0.03

   Strengths

   At $0.14 cached, its GLM-5.3 cached request costs $0.00832, 10% less than the $0.00928 charged by providers at $0.26.¹⁸³

   Limitations

   Its DeepSeek V4.1 Flash 1017 cached rate is $0.007 against $0.006 elsewhere, and Fast costs 1.5× standard on GLM-5.3 ($2.10 / $6.60) and 2× on DeepSeek V4.1 Flash 1017 ($0.60 / $2.40).¹⁸³

   ¹⁸³

3. Fireworks

   The standard rate card on all three current models, at its Standard global tier.¹⁷

   GLM-5.3

   $0.26

   DeepSeek V4.1 Flash 1017

   $0.006

   GLM-5.3-Flash

   $0.03

   Strengths

   With Novita AI, DeepInfra and Together AI, it is one of four providers here that price all four models, including GPT-OSS 120B at $0.15 / $0.60.¹⁷

   Limitations

   US-region pricing is 1.5× and Priority 1.25× the Standard rates shown; GLM-5.3 in the US region costs $2.10 / $6.60.¹⁷

   ¹⁷

4. Novita AI

   The shared standard rate card on all three current models, with no discount on input, output or cached input.^{89101112131821161723}

   GLM-5.3

   $0.26

   DeepSeek V4.1 Flash 1017

   $0.006

   GLM-5.3-Flash

   $0.03

   Strengths

   It prices all four models; on GPT-OSS 120B its $0.05 / $0.25 rates undercut Groq, Together AI, Fireworks and Cerebras.^{89101112131821161723}

   Limitations

   It is not cheapest outright on any model: DeepInfra’s list rates are lower on GLM-5.3 and DeepSeek V4.1 Flash 1017, Baseten’s GLM-5.3 cached rate is $0.14 against Novita AI’s $0.26, and its GPT-OSS 120B has no cached-input rate.^{89101112131821161723}

   ^{89101112131821161723}

5. Together AI

   The standard rate card on the three current models, with GPT-OSS 120B at $0.15 / $0.60.^316891011

   GLM-5.3

   $0.26

   DeepSeek V4.1 Flash 1017

   $0.006

   GLM-5.3-Flash

   $0.03

   Strengths

   Its rates on all three current models are identical to Novita AI’s, so choosing between them comes down to service rather than token price.^316891011

   Limitations

   Its GPT-OSS 120B input rate is 3× Novita AI’s and its output rate 2.4× ($0.15 / $0.60 vs $0.05 / $0.25), with no cached rate listed here.^316891011

   ^316891011

6. Z.ai

   First-party GLM-5.3 and GLM-5.3-Flash at the same rates as the standard card.¹⁹

   GLM-5.3

   $0.26

   DeepSeek V4.1 Flash 1017

   Not listed

   GLM-5.3-Flash

   $0.03

   Strengths

   Its cached-input storage is free for a limited time; cached reads bill at $0.26 on GLM-5.3 and $0.03 on GLM-5.3-Flash.¹⁹

   Limitations

   It prices no other model in this matrix, and the free storage is a promotion that can end.¹⁹

   ¹⁹

7. CoreWeave

   The lowest GPT-OSS 120B rates here, but the highest GLM-5.3-Flash cached-input rate.¹¹⁵³

   GLM-5.3

   Not listed

   DeepSeek V4.1 Flash 1017

   Not listed

   GLM-5.3-Flash

   $0.05

   Strengths

   $0.03 input and $0.17 output give it the cheapest GPT-OSS 120B input in the matrix and tie DeepInfra on output.¹¹⁵³

   Limitations

   Its $0.05 GLM-5.3-Flash cached rate, against $0.03 elsewhere, puts the cached request at $0.00120 versus $0.00104, and it lists neither GLM-5.3 nor DeepSeek V4.1 Flash 1017.¹¹⁵³

   ¹¹⁵³

8. DeepSeek

   First-party DeepSeek V4.1 Flash 1017 with time-of-day pricing.²⁰

   GLM-5.3

   Not listed

   DeepSeek V4.1 Flash 1017

   $0.006

   GLM-5.3-Flash

   Not listed

   Strengths

   Off-peak, all three rates halve: $0.15 input, $0.60 output and $0.003 cached.²⁰

   Limitations

   Peak rates match the standard card, and the off-peak rates apply only to traffic sent in off-peak hours.²⁰

   ²⁰