Guide · Choosing a model
Choosing a Synap model
-
Read the live catalogue
Every price, context limit and capability below comes from this one call. The table on this page is built from it, and re-reads it in your browser.
curl https://synap.linkrra.com/v1/models -
Match the job, not the price
Coding, vision and long context each have one obvious answer. The cheapest id is not always the cheapest answer — see the notes.
coding qwen3-coder-30b vision qwen/qwen3-vl-30b-a3b-instruct long context gemma-4-26b-a4b-it tool-heavy meta/muse-glimmer-30b cheapest gpt-oss-20b -
Or let us choose
One id routes to the cheapest model we serve that can do the job, billed at the price of whichever actually ran.
model = "synap-v1"
Models we run ourselves
On dedicated GPUs we operate. Cheapest first. Every one of them accepts tools and streams. All of them take the batch endpoint at −50% on both directions, within 24 hours.
| Model | Best for | $/1M in | $/1M out | Context | JSON |
|---|---|---|---|---|---|
gpt-oss-20b | Fast + cheap general model | $0.017 | $0.085 | 131,072 | Yes |
gemma-4-26b-a4b-it | Fast general model, long context | $0.04 | $0.21 | 131,072 | Yes |
qwen3-coder-30b | Coding, everyday | $0.0675 | $0.27 | 98,304 | Yes |
qwen/qwen3-30b-a3b-instruct-2507 | General, high-throughput | $0.085 | $0.28 | 98,304 | No |
qwen/qwen3-vl-30b-a3b-instruct | Vision | $0.14 | $0.55 | 65,536 | No |
meta/muse-glimmer-30b | Agentic reasoning + tool use (text) | $0.28 | $1.10 | 131,072 | No |
Models we offer alongside our own
Zero data retention, enforced per request. We do not publish a context figure for these from a live source, so this table does not state one.
| Model | Best for | $/1M in | $/1M out |
|---|---|---|---|
deepseek/deepseek-v4-flash | Fast chat | $0.1654 | $0.3308 |
z-ai/glm-5.3-flash | Multilingual, fast | $0.2848 | $0.3376 |
z-ai/glm-5.3 | Multilingual | $1.5065 | $1.9939 |
deepseek/deepseek-v4-pro | Reasoning, deep | $2.2788 | $4.0512 |
Or let us choose
One id instead of a decision. You are billed for whichever model actually ran.
| Model | What it does |
|---|---|
synap-v1 | auto — routes inside the models we serve on dedicated GPUs we operate and picks the cheapest that can do the job (tools/code/reasoning), billed at the price of whichever model actually ran |
synap-v1-pro | auto — adds the external models we offer to our own: the strongest model per job class (long context, reasoning, multilingual) rather than the cheapest |
team-planner | A named alias. Today it runs deepseek/deepseek-v4-pro, at that model’s price. |
team-worker | A named alias. Today it runs qwen3-coder-30b, at that model’s price. |
USD per 1M tokens, input and output, read from /v1/models — the same figures the gateway bills. Context and parameter support from ?format=openrouter. This page re-reads both when it loads. The free-tier allowance is in the notes below.
Notes, prices and limits
The table above is the live catalogue. The model priced here: qwen3-coder-30b —
$0.0675 in and $0.27 out
per 1M tokens, 98,304-token context.
The cheapest id is not always the cheapest answer. gpt-oss-20b is a reasoning model: its reasoning tokens are billed and count against max_tokens. Measured on this API with max_tokens=20, it spent 17 tokens reasoning and returned content: null with finish_reason: "length". Budget generously, or pick qwen/qwen3-30b-a3b-instruct-2507.
All six models we run ourselves accept tools and tool_choice and stream. JSON mode (response_format, structured_outputs) is declared on three of them, not all six — the JSON column above says which.
Context and parameter support are published for the models we run. For the models we offer alongside our own we have no live source for those, so the second table does not have those columns at all — better an absent column than one filled in from a vendor page.
The batch discount above is a 24-hour completion window, so it is for work that can wait — never for anything a person is sitting in front of.
Prices here are read from /v1/models, which is what the gateway bills, and this page re-reads it in your browser — so a number that changes corrects itself here.
On the free tier: every request counts input and output against 100,000 free tokens a day, and answers stop at 4,096 output tokens. Coding agents send long prompts, so the free tier covers a short session; credits cover the rest.
Field names from this API’s own catalogue: /v1/models and /api/account/free-tier, checked 2026-11-03 20:20 UTC. Prices refresh from /v1/models when this page loads.