Inference
Every model. One key. One balance.
Oddysey holds the accounts with every provider on this page. You hold one deck token — the same one your agents use over MCP — and everything you spend lands on one credit balance with one audit trail. The endpoint speaks the OpenAI protocol, so pointing an existing client at it means changing two strings.
- 01
Claude Sonnet 5
ANTHROPICFRONTIERThe default when the answer matters more than the invoice. Long reasoning, careful drafting, and it says when it is unsure.
oddysey-deep- CONTEXT
- 1M
- IN / 1M
- $2.50
- OUT / 1M
- $12.50
- 02
GPT-5.2
OPENAIFRONTIERStrong general reasoning with a large context. Reach for it when a task spans a lot of material at once.
- CONTEXT
- 400k
- IN / 1M
- $2.20
- OUT / 1M
- $17.50
- 03
Gemini 3.7 Flash
GOOGLEBALANCEDA million tokens of context at a price you can point a batch job at. Good at reading long, dull documents.
- CONTEXT
- 1M
- IN / 1M
- $0.47
- OUT / 1M
- $2.35
- 04
DeepSeek V4 Flash
DEEPSEEKFASTThe cheapest thing here by an order of magnitude. What a watch loop or a summariser should run on.
oddysey-fast- CONTEXT
- 1M
- IN / 1M
- $0.11
- OUT / 1M
- $0.21
- 05
Claude Haiku 4.5
ANTHROPICBALANCEDFollows a fussy instruction closely at a price you can leave running. The default for scheduled agents.
oddysey-balanced- CONTEXT
- 200k
- IN / 1M
- $1.25
- OUT / 1M
- $6.25
- 06
Grok 4.6
XAIFRONTIERFrontier reasoning with unusually cheap output tokens — the pick when a task thinks briefly and then writes a lot.
- CONTEXT
- 500k
- IN / 1M
- $2.50
- OUT / 1M
- $7.50
- 07
GPT-5 Mini
OPENAIFASTThe small OpenAI model. Cheap enough for classification and extraction, familiar enough to drop into existing code.
- CONTEXT
- 400k
- IN / 1M
- $0.32
- OUT / 1M
- $2.50
- 08
Qwen3 Max
QWENBALANCEDAlibaba's flagship. Strong multilingual work, and the best of this list at Chinese and Indonesian.
- CONTEXT
- 262k
- IN / 1M
- $0.98
- OUT / 1M
- $4.88
- 09
Llama 4 Maverick
METAOPEN WEIGHTSOpen weights, a million tokens of context, and no vendor to be locked into. Slower to honour a strict schema.
oddysey-open- CONTEXT
- 1M
- IN / 1M
- $0.25
- OUT / 1M
- $1.00
- 10
Mistral Large 3
MISTRALOPEN WEIGHTSEuropean, open weights, and cheap on output. A solid default for anyone who would rather not send prose to a US hyperscaler.
- CONTEXT
- 262k
- IN / 1M
- $0.63
- OUT / 1M
- $1.88
- 11
Qwen3.6 35B
QWENOPEN WEIGHTSOpen weights with a quarter-million token window, routed first to Darkbloom's network of idle Macs — served there at fp4 rather than the fp8 everyone else runs, so it is cheap in exchange for something real. It thinks before it answers and bills the thinking, so leave max_tokens room.
- CONTEXT
- 262k
- IN / 1M
- $0.13
- OUT / 1M
- $1.19
- 12
Venice Uncensored
VENICEOPEN WEIGHTSA Dolphin fine-tune of Mistral Small that refuses far less than anything else here — fewer refusals, not privacy: the call still runs through Oddysey and its upstream, and every token is still logged to bill you for it.
- CONTEXT
- 128k
- IN / 1M
- $0.25
- OUT / 1M
- $1.13
curl https://oddysey.dev/api/v1/chat/completions \
-H "Authorization: Bearer $ODDYSEY_TOKEN" \
-H "content-type: application/json" \
-d '{
"model": "anthropic/claude-sonnet-5",
"messages": [{ "role": "user", "content": "Is a -5% watch threshold sensible for NVDA?" }]
}'import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://oddysey.dev/api/v1",
apiKey: process.env.ODDYSEY_TOKEN, // odsy_… issued in the Command Deck
});
// Any model id on this page. Streaming, JSON mode and stop sequences all work.
const answer = await client.chat.completions.create({
model: "deepseek/deepseek-v4-flash",
messages: [{ role: "user", content: "Summarise this ledger." }],
});HOW IT IS BILLED
Credit is held before the model runs and reconciled against the tokens it actually used; the difference comes back as a refund in your ledger. New decks start with $0.05 — enough to wire a client up and watch it work.
WHAT THE ALIASES MEAN
oddysey-fast and friends are promises about price and latency, not about which lab built the model. They can be re-pointed when something better and cheaper appears — which is exactly why an agent should store one.
RUNNING OUT
An empty balance answers 402 with x402 payment requirements attached, so an agent with a wallet can settle and retry on its own. A person tops up in the deck instead.
Model names and marks belong to their respective owners and are shown to identify what you are calling. Oddysey resells access; it does not train, host, or speak for any of them. Prices are Oddysey list prices and include the routing margin — nothing is added at the invoice.