LimitWatch

Home › Compare

AI model prices: aggregators vs buying direct

Per-million-token list prices (USD, standard tier, text) for representative models, bought from the model maker versus through OpenRouter, Hugging Face Inference Providers and Vercel AI Gateway. Aggregator prices come from each one's official machine-readable model list, refreshed at every build; direct prices are read from our stored snapshots of official pricing pages. Every number shows its source and fetch date. If we could not fetch or verify a price, it shows n/a; we never fill one in. Percent difference = (route − direct) ÷ direct.

ModelRouteInput $/1M tokensOutput $/1M tokensvs direct (in / out)Source
GPT-4oDirectn/a†n/a†n/aGPT-4o text pricing is not in our stored snapshot of OpenAI's pricing page (the openai.com pricing page blocks automated access).
GPT-4oOpenRouter$2.50$10.00no direct priceAPI, fetched 2026-10-23
GPT-4oVercel AI Gateway$2.50$10.00no direct priceAPI, fetched 2026-10-23
GPT-4oHugging Face Inference Providersn/a†n/a†n/amodel not listed in the official price list fetched 2026-10-23
Claude Sonnet 5Direct (Anthropic (Claude))$2.00$10.00baselinesnapshot, fetched 2026-10-23
Claude Sonnet 5OpenRouter$2.00$10.00+0% / +0%API, fetched 2026-10-23
Claude Sonnet 5Vercel AI Gateway$2.00$10.00+0% / +0%API, fetched 2026-10-23
Claude Sonnet 5Hugging Face Inference Providersn/a†n/a†n/amodel not listed in the official price list fetched 2026-10-23
Claude Sonnet 5.5Direct (Anthropic (Claude))$2.00$10.00baselinesnapshot, fetched 2026-10-23
Claude Sonnet 5.5OpenRouter$2.00$10.00+0% / +0%API, fetched 2026-10-23
Claude Sonnet 5.5Vercel AI Gateway$2.00$10.00+0% / +0%API, fetched 2026-10-23
Claude Sonnet 5.5Hugging Face Inference Providersn/a†n/a†n/amodel not listed in the official price list fetched 2026-10-23
Llama 3.3 70B InstructDirectn/a†n/a†n/aMeta does not sell Llama 3.3 70B through a first-party API that we track; no direct price.
Llama 3.3 70B InstructOpenRouter$0.10$0.32no direct priceAPI, fetched 2026-10-23
Llama 3.3 70B InstructHugging Face (novita)$0.14$0.40no direct priceAPI, fetched 2026-10-23
Llama 3.3 70B InstructHugging Face (scaleway)$1.03$1.03no direct priceAPI, fetched 2026-10-23
Llama 3.3 70B InstructHugging Face (ovhcloud)$0.74$0.74no direct priceAPI, fetched 2026-10-23
Llama 3.3 70B InstructVercel AI Gateway$0.72$0.72no direct priceAPI, fetched 2026-10-23
Gemini 2.5 FlashDirect (Google (Gemini))$0.30$2.50baselinesnapshot, fetched 2026-10-23
Gemini 2.5 FlashOpenRouter$0.30$2.50+0% / +0%API, fetched 2026-10-23
Gemini 2.5 FlashVercel AI Gateway$0.30$2.50+0% / +0%API, fetched 2026-10-23
Gemini 2.5 FlashHugging Face Inference Providersn/a†n/a†n/amodel not listed in the official price list fetched 2026-10-23

Cheapest verified route per model

Based on input + output list price only, ignoring fees, caching, batch and regional pricing. Ties are listed together.

Platform and credit fees (official wording)

We pass through the pricing of the underlying model providers without any markup, so you pay the same rate as you would directly with the provider.

Source: OpenRouter FAQ, last checked 2026-10-22.

OpenRouter charges a fee when you purchase credits.

Source: OpenRouter FAQ, last checked 2026-10-22.

OpenRouter's credit-purchase fee percentage: not stated in our fetched copy of the FAQ (the number did not render). Not included in the table.

Usage above the allowance has a fee of 5 % of what the same model and provider would normally cost on OpenRouter.

Source: OpenRouter FAQ (BYOK), last checked 2026-10-22.

Hugging Face charges you the same rates as the provider, with no additional fees.

Source: Hugging Face Inference Providers pricing, last checked 2026-10-23.

AI Gateway charges no markup and no platform fee on tokens.

Source: Vercel AI Gateway pricing, last checked 2026-10-23.

You're responsible for any payment processing fees that may apply.

Source: Vercel AI Gateway pricing, last checked 2026-10-23.

Poe and T3 Chat

Poe: not comparable. poe.com's robots.txt disallows our crawler for its pricing/about page, so we did not fetch it and cannot derive a per-token cost from official numbers.

T3 Chat: not comparable. Its site returned HTTP 429 (rate limited) to our crawler; we did not retry around that, and we have no official per-token numbers to derive an effective cost.

Price-list refresh at this build: openrouter: refreshed, huggingface: refreshed, vercel: refreshed. See also how aggregators route requests, price drops & deals, the changelog.