image2.5 from $0.0085 per 1K image · Seedance 2.5 from $0.0961/sec · cached LLM input from $0.40/M — one OpenAI-compatible endpoint at
https://api.apimart.ai/v1, no monthly plan required. (observed 2026-09-17)
Get an API key · Live pricing · Model page
Why teams route through APIMart
- One key, entire catalog. The same
https://api.apimart.ai/v1base URL andAuthorizationheader reach the whole catalog behind one key and 300+ other image, video and language models — switch themodelfield, not your client. - $1 minimum, pay as you go. No subscription and no prepaid plan to size up front: top up from $1 and spend it on calls. There is no free quota to burn through first, so the price in this table is the price you pay.
- The charge comes back in the response. Every call reports the amount billed (
cost/credits_cost), so a spend number is read per call instead of guessed at month end. - Async by design. Submit, take the
task_id, pollGET /v1/tasks/{id}— batching and retries are ordinary queue work, not a bespoke integration.
LLM gateway comparison without the marketing table: pick the dimensions that matter to you, weight them, and score the four gateway archetypes (self-hosted open source, managed control plane, relayed per-unit route, direct vendor API) with a script you can rerun when your requirements change.
Why archetypes instead of a vendor scorecard
Vendor feature matrices go stale in weeks, and a comparison written by an interested party is not a comparison. This repository keeps the part that stays true: the dimensions on which gateways differ, plus a scoring script so the weighting is yours. Named products appear only as examples of an archetype.
python tools/compare.py --show # the criteria matrix
python tools/compare.py --weight control=3 --weight cost_transparency=3 \
--weight setup_effort=2 --weight ops_burden=1
The dimensions that actually differ
| Dimension | Question it answers | Why it decides projects |
|---|---|---|
| Credential model | How many keys does the caller manage? | Key sprawl is the top operational cost of multi-vendor work |
| Response normalisation | Is the envelope the same across providers? | Parser drift after an API revision is a silent outage |
| Billing unit | Which unit do you pay in? | Per-image/per-second makes cost a multiplication; tokens require measurement |
| Retry safety | Is there an idempotency mechanism? | Without it, a timeout plus a retry bills twice |
| Timeout control | Can you set request and job TTLs? | Long generation jobs need a TTL you own |
| Observability | What is exposed per request? | Per-task cost, progress and status are the minimum for reconciling a bill |
| Self-hostable | Can it run in your infrastructure? | Decides data residency and egress review |
| Ops burden | Who runs upgrades and incidents? | The hidden line item in every "free" gateway |
Run your own comparison in one afternoon
- Freeze a request set. 20–50 real prompts covering every surface you use (chat, streaming, tools, images, vision).
- Score the archetypes with
tools/compare.pyusing weights your team agrees on — record who set which weight. - Test two candidates for real. Same prompts, same acceptance rules; record p50/p95 latency, error classes, retry behaviour and cost per accepted artefact.
- Check the boring parts: key rotation, quota exhaustion behaviour, retention and region, and what happens to an in-flight job when you fail over.
- Only then decide, and keep the rollback as a configuration change rather than a code change.
What this repository is not
- It is not a ranking of vendors, and it does not claim that any archetype wins: a two-person team and a platform team with an SRE rotation should reach different conclusions from the same matrix.
- It is not a benchmark. Latency and price numbers belong in your own run, on your own prompts.
What that costs at scale
| Workload | Cost at the observed rates |
|---|---|
| 1,000 GPT Image 2.5 renders (1K) | $8.50 |
| 10 minutes of Seedance 2.5 at 480P (600s) | $57.66 |
| 1M cached LLM input tokens | from $0.40 |
Linear at the observed per-unit rate, no volume discount assumed. Snapshot 2026-09-17; re-check the live table before committing a budget.
First-call troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
401 / invalid api key |
key missing, truncated, or a stray newline pasted into the header | Re-copy it from the console; the header is Authorization: Bearer $APIMART_API_KEY |
| balance / credit error | the account has no balance | Top up from $1 in the console — there is no free quota to fall back on |
429 |
concurrent requests on one key | Back off, then retry the same request with the same Idempotency-Key |
400 / model not found |
wrong route for the id: the per-unit alias needs its version, the official id must not send one |
Copy the exact model value from the route table above |
task ends failed |
prompt rejected by the filter, or a reference image URL expired | Re-submit with a new Idempotency-Key and re-host the reference image |
| result URL stops working | result links expire | Download the file as soon as the task reports completed |
Related searches
llm gateway comparisonai api gatewayllm gateway open sourcellm gateway vs proxyai api aggregatoropenai compatible apiai api pricing comparison
Start with $1. Get an API key → check live pricing → open the whole catalog behind one key in the model library. The first call is three steps: submit, poll task_id, read the charged amount off the response.
Attributed links (how this repository is measured)
| Purpose | Attributed link | Target |
|---|---|---|
| Browse the model catalog | https://go.apimart.ai/k-dda2cc | apimart.ai/model |
| Current pricing page | https://go.apimart.ai/k-9f4ab7 | apimart.ai/pricing |
| Get an API key | https://go.apimart.ai/k-95b075 | apimart.ai/keys |
Outbound APIMart links are minted through the promo link API; hand-made tracking parameters are rejected by
tools/check_links.py in CI.
Disclosure
This repository describes gateway archetypes and a scoring method. It is published to document that method, not to claim official status for any vendor; product names appear as examples and belong to their owners. Verify each cell against the vendor's documentation for your own account.
Repository map
README.md comparison method, dimensions and workflow
data/gateways.json archetypes, dimension values and scores (schema: llm-gateway-comparison-v1)
tools/compare.py weighting and scoring
tools/check_links.py attribution guard (CI)
Comments