Pricing articles
Updated
What LLM API usage really costs: per-token math, flat-rate plans, cached input and break-even points, all from published prices. 20 articles, each opening with a direct answer. Every outside fact links to its source with the date we checked it.
LLM API price changes in October 2026: 3 rises, no cuts
Few, and no cuts. Comparing archived pricing pages of 7 LLM API providers from early October to early November 2026, we found 3 list-price rises: two at DeepInfra (Mistral Nemo, Gemma 4 31B) and one at Together AI, where Qwen3.8 Flash went back to its September price. 9 models were newly priced and 11 left the pages.
Do LLM API prices only go down? 52 rises and how to budget for them
No. In archived pricing pages from 11 providers, we found 52 list-price rises on the same model between 2024 and November 2026, against 100 cuts. The median rise was 40% on a blended price, and 16 rises doubled it or more. Most were at DeepInfra (40) and Together AI (7), but Anthropic, Mistral and DeepSeek raised prices too.
How often do LLM API prices change? 3 years of archived pricing pages
Rarely, for most models. In 308 archived pricing pages from 11 providers (January 2023 to November 2026), 398 of 498 models seen in more than one month (80%) never changed price. We found 161 list-price changes: 100 cuts, 52 rises and 9 mixed. DeepInfra made 121 of them. OpenAI, Anthropic and Google repriced 8 of 126 models between them.
How to reduce LLM API costs: a 10-step checklist, in order
To reduce LLM API costs, first log tokens per request, then work in this order: keep the prompt prefix stable so it caches, trim prompts, cap output and reasoning, use the smallest model that passes your tests, batch or shift work that can wait, cache repeated answers, bound retries, and move to a flat plan once usage passes break-even.
LLM API price cuts 2023–2026: the biggest drops, with sources
The largest list-price cuts in our archive of 11 providers' pricing pages were 80% on OpenAI o3 (mid-2025) and Mistral Codestral (late 2024), 75% on Gemini 1.5 Flash and 81% on Mistral Small 3 at Together AI. Makers' cuts were large (median 58%); hosts' were small (median 14%). We also found 52 rises, including Claude 3.5 Haiku.
Open-weight vs closed model API cost: what the price data shows
Usually, but not by model name alone. On prices checked on 15 November 2026, GLM 5.3 at its maker costs less per token than GPT-6.1 Sol or Claude Sonnet 5.5, yet small closed models such as GPT-6 Luna cost about the same as hosted open models. The host also matters: one open model can cost several times more at one host than another.
10 ways to cut coding agent LLM costs, with estimated savings
Most spend is input resent every turn. Keep the prompt prefix stable so it caches, name files, trim tool output, lower reasoning effort, cap loops and retries, and route easy steps to cheaper models. On DeepSeek V4.1 Flash, cached input costs about 3% of new input. Past about 917 million input tokens a month, a $60 flat plan is cheaper.
A weekly LLM plan for short projects: Orbit vs hour vs month
Orbit is one week of unlimited DeepSeek V4.1 Flash for $15 per session, paid from your BroadNet balance, with no token meter and one request at a time. It fits a sprint, hackathon or migration. For an afternoon, Meteor at $1 an hour is cheaper; for continuous use, Full Moon at $60 beats back-to-back weeks ($64.29 over 30 days).
Cached input tokens: how to cut an LLM API bill
Cached input tokens are prompt tokens the provider has already processed and can reuse, so they are billed at a lower rate. On BroadNet Inference, DeepSeek V4.1 Flash bills cached input at $0.0055 per 1M tokens against $0.175 for new input. Keep the stable part of every prompt first and identical between calls to get the discount.
Cheapest vision LLM API: cost per image and per page, worked out
A vision LLM bills an image as input tokens, so 1,000 images of about 1,000 tokens each cost the input price per 1M tokens. On BroadNet the cheapest is Gemma 4 VL 26B at $0.10 input and $0.22 output per 1M. DeepInfra lists Google's Gemma 4 26B A4B with lower input and higher output prices.
DeepSeek V4.1 Flash API pricing: per token, cached and plans
On BroadNet Inference, DeepSeek V4.1 Flash costs $0.175 per 1M input tokens, $0.0055 per 1M cached input tokens and $0.52 per 1M output tokens, reasoning included. Or pay a flat price per session: $60 a month, $15 a week or $1 an hour, with no token meter, one request at a time.
Flat-rate vs pay-per-token LLM API: the break-even math
For DeepSeek V4.1 Flash on BroadNet Inference, a flat Full Moon plan ($60 a month) beats pay-per-token at roughly 917 million input and 46 million output tokens a month, assuming 80% cache hits. A full-time coding agent usually crosses that line. A side project or chat assistant usually does not.
Free LLM API credit: how far a starting credit really goes
New BroadNet Inference accounts get $5 of free pay-as-you-go credit. It works on BroadNet Gemma 4 VL 26B ($0.10 per 1M input tokens), the transcription models and hourly keys. That is tens of millions of Gemma tokens, or about 1,190 minutes of Arabic transcription. DeepSeek V4.1 Flash, GLM 5.3 and Qwen3.8 27B open after your first top-up.
Hourly LLM API key: a flat price per hour, explained
An hourly key is a prepaid API key that works for one hour at a flat price. On BroadNet Inference it costs $0.50 for one hour of Gemma 4 VL 26B: one stream, 120 requests a minute and 300,000 tokens a minute, with no token meter. It suits demos, workshops and short batch jobs with a fixed cost.
How to estimate LLM API cost before you launch a feature
Estimate LLM API cost by modelling one user action, not one API call: count calls per action, measure input, cached and output tokens per call on real samples, then multiply by users and actions per month. Add reasoning and retries, and price low, expected and high scenarios. Our worked example on DeepSeek V4.1 Flash comes to $42.67 a month.
LLM API pricing explained: input, cached, output and reasoning
LLM APIs bill per million tokens, with separate prices for input, cached input and output. Reasoning tokens are billed as output, images count as input tokens, and speech to text is billed per audio minute. On BroadNet Inference, DeepSeek V4.1 Flash costs $0.175 input, $0.0055 cached and $0.52 output per 1M tokens.
LLM API spend control: budgets, key limits and loop protection
Control LLM API spend in layers: a prepaid balance as the account ceiling, one key per app locked to your server IPs, max_tokens and step caps on every request, and an in-code budget that stops a job before it overspends. Log tokens per key and alert on your own numbers. Flat plans and hourly keys cap the worst case.
LLM cost per user: unit economics for a SaaS AI feature
LLM cost per user = requests per active user per month × tokens per request × the per-1M token prices. Measure it per feature from the usage field, price tiers on the 90th-percentile user rather than the average, and add fair-use limits for the tail. Cheaper models, caching and low reasoning effort cut the number directly.
Reasoning tokens: why you pay for more output than you see
A reasoning model writes hidden thinking before its answer, and those reasoning tokens are billed at the output price. A 200-token reply can carry 1,000 or more tokens of thinking. On BroadNet Inference, DeepSeek V4.1 Flash, GLM 5.3 and Qwen3.8 27B reason. Lower the cost with reasoning_effort set to low, a sensible max_tokens, and a non-reasoning model for simple steps.
Tokens per word in Arabic vs English: why Arabic costs more
There is no single figure: it depends on each model's tokenizer. In BroadRouter's measured dataset, the same Modern Standard Arabic text needs 1.31× (Qwen3.8 27B) to 1.77× (GLM 5.3) the tokens of its English translation, and 1,000 words of English prose come to about 1,290 tokens on average. Measure your own ratio from usage.prompt_tokens.





