Gateway

Models & providers

26 models you can ask for today across 9 providers, out of 45 keys in the registry — the other 19 are previous-generation keys kept so older configurations still price. Generated from the registry the router and the meter both read, so it cannot drift from what the Gateway actually serves.

A Lobstack model key is what you send as model. It maps to a provider, a provider-native model id, a capability tier and a price per million tokens. The same registry answers three questions: which model the router may pick, which id goes on the wire, and what the tokens cost.

Prices are list prices in USD per million tokens, used for metering rather than for invoicing. Every one of them was last checked against the provider's own pricing page on 2026-09-08. Prices in this category move often; check again before you quote one.

Current generation

These are the 26 keys a picker should offer first, and the pool the router draws from for auto.

Key	Provider	Tier	Context	$/M in	$/M out
claude-fable-5-1	Anthropic	flagship	1M	$10.00	$50.00
gpt-6-astra	OpenAI	flagship	1.05M	$10.00	$50.00
claude-opus-5	Anthropic	flagship	1M	$5.00	$25.00
gpt-5.6	OpenAI	flagship	1.05M	$4.00	$20.00
deepseek-v4-pro	DeepSeek	flagship	1M	$1.10 *	$4.40 *
kimi-k3	Moonshot	premium	1.05M	$3.00	$15.00
gemini-3.1-pro	Google	premium	1.05M	$2.00	$12.00
gpt-5.6-terra	OpenAI	premium	1.05M	$2.00	$12.00
grok-4.6	xAI	premium	500K	$2.00	$6.00
grok-4.5	xAI	premium	500K	$2.00	$6.00
qwen3.8-max	Alibaba	premium	1M	$2.00	$6.00
mistral-medium-3.5	Mistral	premium	256K	$1.50	$7.50
claude-sonnet-5	Anthropic	standard	1M	$2.00	$10.00
grok-4.3	xAI	standard	1M	$1.25	$2.50
gemini-3.8-flash	Google	standard	1.05M	$0.75	$3.75
qwen3.7-plus	Alibaba	standard	1M	$0.40	$1.60
deepseek-v4-flash	DeepSeek	standard	1M	$0.27 *	$1.10 *
claude-haiku-4-5	Anthropic	small	200K	$1.00	$5.00
grok-build-0.1	xAI	small	256K	$1.00	$2.00
gpt-oss-120b	Groq	small	131K	$0.15	$0.60
mistral-small-4	Mistral	small	256K	$0.15	$0.60
qwen3.8-flash	Alibaba	small	1M	$0.15	$0.47
gemini-3.1-flash-lite	Google	nano	1.05M	$0.25	$1.50
gpt-5.6-luna	OpenAI	nano	1.05M	$0.20	$1.20
ministral-8b	Mistral	nano	256K	$0.15	$0.15
gpt-oss-20b	Groq	nano	131K	$0.075	$0.30

Previous generation

These 19 keys are kept so stored configurations keep resolving. Where a provider retired a model outright, the key survives and points at the documented successor — and it is priced at the successor's rate, because a retired model now served by a cheaper engine must not keep billing at the old one.

5 of them redirect today: Grok 3 → Grok 4.3; Grok 4.1 Fast → Grok 4.3; DeepSeek R1 → V4 Pro; DeepSeek V3 → V4 Flash; Llama 3.3 70B → GPT-OSS 120B.

Key	Provider	Tier	Context	$/M in	$/M out
claude-opus-4-6	Anthropic	flagship	1M	$5.00	$25.00
gpt-5.2	OpenAI	flagship	256K	$5.00	$20.00
gpt-5	OpenAI	premium	256K	$5.00	$15.00
qwen3.7-max	Alibaba	premium	1M	$2.50	$7.50
qwen-max	Alibaba	premium	256K	$1.20	$6.00
deepseek-r1	DeepSeek	premium	1M	$1.10 *	$4.40 *
claude-sonnet-4-5	Anthropic	standard	200K	$3.00	$15.00
gpt-4o	OpenAI	standard	128K	$2.50	$10.00
gemini-3.5-flash	Google	standard	1.05M	$1.50	$9.00
gemini-2.5-pro	Google	standard	1.05M	$1.25	$10.00
grok-3	xAI	standard	1M	$1.25	$2.50
mistral-large	Mistral	standard	256K	$0.50	$1.50
qwen-plus	Alibaba	standard	1M	$0.40	$1.20
gpt-5-mini	OpenAI	small	256K	$0.50	$2.00
deepseek-v3	DeepSeek	small	1M	$0.27 *	$1.10 *
mistral-small	Mistral	small	256K	$0.15	$0.60
llama-3.3-70b	Groq	small	131K	$0.15	$0.60
grok-4.1-fast	xAI	nano	1M	$1.25	$2.50
gpt-5-nano	OpenAI	nano	128K	$0.15	$0.60

Unverified prices

4 keys are marked with an asterisk above: deepseek-v4-pro, deepseek-v4-flash, deepseek-r1, deepseek-v3. DeepSeek publishes list prices in CNY and its English pricing page has been serving the wrong document, so these USD figures are inherited from the previous catalog and could not be re-confirmed on 2026-09-08.

They are not currency-converted, because DeepSeek's USD prices have historically not tracked a straight FX conversion, and inventing a rate would produce a number that looks checked and is not. The flag is carried in the registry as priceUnverified and surfaced here rather than hidden: a figure we cannot source is not one you should have to take on faith. Metering still uses these numbers, so cost reported for a DeepSeek request is an estimate against an unconfirmed list price.

What is not here

Anthropic Mythos

Mythos exists and it sits above Fable, and it is deliberately absent from this catalog. It ships through trusted-access programmes and is not generally available on the API. Listing it would advertise a model no customer of ours can call.

Groq's Llama models

Groq no longer publishes a public price for its Llama models, so a Llama request can no longer be metered honestly. Rather than delete the key and break stored configurations, the llama-3.3-70b key survives and points at openai/gpt-oss-120b, the Groq-hosted model that does have a published price. A request naming Llama is served by GPT-OSS 120B and priced as GPT-OSS 120B. x-lobstack-model reports the key you asked for, so read the label in this table to know what actually ran.

Aliases

The resolver accepts a set of non-canonical spellings that turn up in stored configurations: provider-native ids, hyphenated variants, and models listed before this catalog existed. gemini-2-5-pro resolves to gemini-2.5-pro; gpt-4o-mini resolves to gpt-5-mini; kimi-k2 resolves to kimi-k3.

Aliases matter for billing, not just for routing. An unresolved key prices at zero, which is the most expensive way to be wrong here: a live agent ran on gemini-2-5-pro and every one of its requests metered free. Pricing goes through the same resolver as routing so that cannot happen again. Resolving a key for routing but not for pricing would be worse than not resolving it at all.

Listing models over the API

GET /api/gateway/v1/modelsbash

curl https://www.lobstack.ai/api/gateway/v1/models

Response (one entry)json

{
  "id": "claude-sonnet-5",
  "object": "model",
  "owned_by": "anthropic",
  "label": "Claude Sonnet 5",
  "tier": "standard",
  "context_window": 1000000,
  "price_per_mtok": { "input": 2, "output": 10 },
  "managed": true
}

The listing is OpenAI-shaped with Lobstack fields added, so client.models.list() works. It returns the canonical keys only; aliases resolve on a request but are not enumerated.

Serving a key needs a configured provider keyIn managed mode the router only picks models whose provider has a key configured on the deployment, so a key in this table may not be reachable on a given install. The route-preview endpoint reports managed_key_configured per model so the difference is visible rather than surprising. In BYOK mode routing never leaves your key's own provider.