LLM API pricing comparison: the whole catalog, sorted by input price

How to read this table

This LLM API pricing comparison covers every per-token rate in the catalog in one place, so you can rank models by what they actually charge instead of by a headline number.

• All rates are USD per 1,000,000 tokens, from the live pricing page data.
• 104 of the 126 catalog entries publish both an input and an output rate per 1M tokens; the remaining 22 are image and video generation models billed per generated image or second, which a per-token rate cannot express.
• Access tier shows which groups can call the model. In the current catalog the group multiplier is 1 for all four groups, so the tier decides what you can reach, not what you pay per token.
• Click a column header in the table to re-sort it. The default order is input price, lowest first.

All 104 per-token rates, sorted by input price

Rates read from FlowBarAI's pricing API on September 12, 2026. Prices are subject to change; the authoritative source is the pricing page.

Which access tier reaches how much of the catalog

Access tiers do not change the per-token rate in the current catalog — the group multiplier is 1 for all four groups — but they decide which models you can call at all. The counts below are catalog entries reachable per group, read from the same pricing data.

What the numbers show

• Input rates span $0 to $142.5 per 1M tokens (openai/o1-pro is the most expensive input rate in the catalog).
• Output rates span $0 to $570 per 1M tokens.
• 10 entries are priced at or below $0.10 per 1M input tokens, and 73 at or below $1.00 per 1M input tokens.
• Output costs more than input almost everywhere: the median output rate is 4× the input rate across the 102 entries where both are non-zero.
• The cheapest paid per-token entry is Qwen/Qwen3-Coder-30B-A3B-Instruct at $0.0665 per 1M input tokens and $0.266 per 1M output tokens.

Models billed outside per-token pricing

These 22 catalog entries are image and video generation models. They publish no per-1M-token rate, so they are listed here by name instead of being given a price this page cannot support. Per-generation rates are on the pricing page.

• doubao-seedream-4.0 ($10+ · $30+ · $80+)
• doubao-seedream-4.5 ($10+ · $30+ · $80+)
• Doubao-seedream-5.0-lite ($10+ · $30+ · $80+)
• doubao-seedream-5.0-pro ($10+ · $30+ · $80+)
• happyhorse-1.0-i2v ($10+ · $30+ · $80+)
• happyhorse-1.0-r2v-20260618 ($10+ · $30+ · $80+)
• happyhorse-1.0-t2v ($10+ · $30+ · $80+)
• happyhorse-1.1-i2v ($10+ · $30+ · $80+)
• happyhorse-1.1-r2v ($10+ · $30+ · $80+)
• happyhorse-1.1-t2v ($10+ · $30+ · $80+)
• qwen-image-edit-max ($10+ · $30+ · $80+)
• qwen-image-edit-plus ($10+ · $30+ · $80+)
• wan2.2-kf2v-flash ($10+ · $30+ · $80+)
• wan2.6-r2v ($10+ · $30+ · $80+)
• wan2.6-r2v-flash ($10+ · $30+ · $80+)
• wan2.7-i2v ($10+ · $30+ · $80+)
• wan2.7-image ($10+ · $30+ · $80+)
• wan2.7-image-pro ($10+ · $30+ · $80+)
• wan2.7-r2v ($10+ · $30+ · $80+)
• wan2.7-t2v ($10+ · $30+ · $80+)
• wan3.0-video ($10+ · $30+ · $80+)
• wan3.0-video-prime ($10+ · $30+ · $80+)

How to compare these rates with another provider

The same method is written up in more detail in how to compare LLM API pricing, with cost-saving reasoning in how to choose a cheap LLM API. The catalog behind this table is browsable on the models page, and endpoint details are in the docs.

1. Convert both sides to the same unit. Per-1,000 and per-1,000,000 quotes are mixed constantly, and a unit mistake silently changes a ranking by 1,000×.
2. Split your workload into input tokens and output tokens, then multiply each by its own rate. The output multiple is large enough that a blended number misleads.
3. Check the access tier you actually qualify for, because a model you cannot call is not a cheaper option.
4. Add the costs outside token pricing: cache reads and writes, image and video generation, and any subscription or minimum spend.
5. Confirm the failure policy. Failed requests are not charged on FlowBarAI; check each other provider's terms.

Frequently asked questions

Which LLM API is cheapest per token?

In the pricing data read on September 12, 2026, 2 entries are listed at $0 per 1M input and output tokens. Among paid entries the lowest input rate is $0.0665 per 1M tokens (Qwen/Qwen3-Coder-30B-A3B-Instruct), but the cheapest choice still depends on your output volume and the tier you can access.

Are these prices current?

They were read from FlowBarAI's pricing API on September 12, 2026 and are subject to change. The pricing page is the authoritative, always-current source, and the API is the machine-readable one.

Why do output rates differ so much from input rates?

In this catalog the median output rate is 4× the input rate across the 102 entries where both are non-zero. Output-heavy workloads are therefore dominated by the output rate, while summarization and classification are dominated by input.

Does my top-up tier change the per-token rate?

No. In the current catalog the group multiplier is 1 for all four access groups, so the tier decides which models you can reach rather than the rate per token. Cumulative successful top-ups of $10, $30 and $80 widen that reach.

How many models are in the catalog?

126 entries as read on September 12, 2026: 104 with published per-1M-token rates and 22 image or video generation models billed per generated unit.

What happens if a request fails?

Failed requests are not charged on FlowBarAI.

Related reading

• OpenRouter alternatives: multi-model and self-hosted options
• Free LLM API access: what a real free trial includes
• LiteLLM alternatives: self-hosted vs managed
• How to choose a cheap LLM API
• How to compare LLM API pricing
• AI video generation platform workflow
• FlowBarAI in English: gateway overview