Skip to content

Every agent's model. One place. Codex on DeepSeek, Claude Code on Kimi, from the menu bar.

provider: a model is priced at its maker's own API before a reseller's (bestony on Discord)

yetone65bd4e4f7241
diff --git a/docs/subsystems/providers-accounts.md b/docs/subsystems/providers-accounts.md
index d3beec4c8..78bc1b3c5 100644
--- a/docs/subsystems/providers-accounts.md
+++ b/docs/subsystems/providers-accounts.md
@@ -53,6 +53,7 @@ keys and accounts per request is in [Gateway routing and fallback](gateway-routi
| A remote magpie's service tiers | A remote magpie's `/v1/models` gives each model the `service_tiers` its own Codex is offered on it (`codexTiers` → `codexcat.ServiceTiers`, from the same `CodexCatalog` and `codexcat.Entries` as its `/v1/codex/models`: Fast on a ChatGPT account's GPT model, the tiers of Codex's own entry on a relay's, none on a model it offers none). The local keeps the ids (`catalog.Model.Tiers` → `Entry.Tiers`), `codexListed` hands them on for the remote's models, and `codexcat.Entries` offers them to the local Codex with Codex's own names (`namedTiers`). The gateway sends the tier Codex asks for to a remote magpie as it was asked (`Request.OwnTier`, as for a provider added by its address; dropped and not sent again if refused), and the remote serves it as its own Codex's turn (#1234). An agent's picker offers the bolt on a remote's model whose tiers hold `priority` (`provider.CanFast` reads them with `RemoteTiers`; Ultrafast too where they hold `ultrafast`, `UltraOn`), and a fast pick of one goes on with `service_tier` `priority` on the remote's Chat or Responses: an agent on Messages, whose fast is Claude's `speed`, has it turned into the tier the remote offers (`forwardTranslated`), so Claude Code's pick reaches the remote's vendor too (littlekk88 on Discord; `TestRemoteMagpieFastPick`, red without either half). A remote from before this sends no tiers, and its models are offered none, as before. The list gives every client the tiers, not only a magpie asking with `X-Magpie-Drawers`: Cindy shows its Fast switch on a model whose `service_tiers` hold `priority` (#1914, `TestModelsListTiersForEveryClient`). | [`codexcat.go`](../../internal/codexcat/codexcat.go), [`codex_models.go`](../../internal/provider/codex_models.go), [`codex_catalog.go`](../../internal/gateway/codex_catalog.go), [`gateway.go`](../../internal/gateway/gateway.go), [`live.go`](../../internal/catalog/live.go) |
| A relay's Codex entries | A provider the user added by its address with a Responses URL may list its models in Codex's own shape (`slug`, `supported_reasoning_levels`, `shell_type`, `base_instructions`: `catalog.CodexEntry`), plainly or when asked with Codex's `client_version`, as sub2api does (ctopsc on Discord: GPT-6.1 Terra was handed models.dev's 922000 window where the relay said 272000). `fetchOne` keeps such a row whole (`catalog.Model.Codex`, with its window, max window and inputs read off it), and when the plain list has none it asks the same URL once more with `client_version` (`codexEntries`): only where the list is fetched (add, key change, Refresh), never per Codex poll. The answer only adds entries to models the plain list has; one that fails or can't be read keeps the last list's entries (failed read = unknown), one that answers plainly says there are none. `codexcat.listedEntry` hands Codex the relay's entry as given, `base_instructions` included, with magpie's slug, name, priority and " via magpie" description; the user's window, compaction, levels and image setting still win (levels keep the relay's descriptions). A saved list from before is enriched at its next Refresh. `TestCodexListedEntry`, `TestCodexListedEntryInPlainList`. | [`codex_entries.go`](../../internal/provider/codex_entries.go), [`live.go`](../../internal/catalog/live.go), [`codexcat.go`](../../internal/codexcat/codexcat.go) |
| An endpoint's catalog | A key's provider with no catalog of its own (none named, none from a preset) is priced from the models.dev catalog whose `api` is one of its endpoints (`EndpointCatalog`, `catalog.AtEndpoint`; a trailing slash and case don't count, and an Anthropic base, kept as its root, is also matched with its `/v1`). Only when exactly one catalog is at that address, and it isn't a plan's (an id with `-plan`, or a `membershipCatalogs` one): a plan lists its membership's $0, and two catalogs at one address can't be told apart. A subscription's account never gets one. A provider at AMD's own endpoint is priced from `amd`, which lists Qwen3.8-27B at $0, a price, where its maker sells no such model and it had none (#2052); a relay's endpoint matches nothing and its model stays unpriced, never priced at a reseller's guess. `magpie model price` names the catalog, and says how to give a price when none is known. | [`models.go`](../../internal/provider/models.go), [`catalog.go`](../../internal/catalog/catalog.go) |
+| The maker's own price | A model priced at its maker (`MakerPrice`, a provider's `ListPrice` after its own catalog, `PricedName`; through `makerListPrice`) is priced at a models.dev provider that lists it as its own (`catalog.MakerPriced`): one whose entry's `canonical_model_id` names it as the maker, or whose id is the maker's with a region (`alibaba-cn`), never a plan's (`-plan`) and never at $0. A maker preset that is such an API is asked first, in preset order, then the other owners by id; only then the maker presets as before (`PricedBy`). A maker preset that resells another lab's model no longer prices it: Mistral's `deepseek-v4-1-flash-1017` at $0.30/$1.20 gave way to DeepSeek's $0.15/$0.60, and a lab that isn't a preset (Meta's `muse-spark`) is priced at all. So a subscription's model (Command Code's, bestony on Discord) is estimated at what its maker charges (`TestMakerPriceIsTheMakersOwn`). A model no maker lists (Kimi K2.5, GLM-5.2 Fast) stays unpriced; the user gives it a price or a same-as model. | [`models.go`](../../internal/provider/models.go), [`catalog.go`](../../internal/catalog/catalog.go) |
| A membership's $0 catalog | models.dev lists every model of Kimi Code's plans (`kimi-code-plan-global`, `kimi-code-plan-cn`) and of Xiaomi MiMo's Token Plans (`xiaomi-token-plan-cn`, `-sgp`, `-ams`) at $0: what the membership charges per token, not what the model costs. `membershipCatalogs` are never read for a price (`priceCatalogs`, `makerPriceCatalogs`), so a Xiaomi MiMo model is priced at `xiaomi`'s list price at every region, and each region of the preset names its own catalog, pay as you go `xiaomi` alone (#1873). A Kimi Code provider's model is priced at the Kimi API model it is (`kimiCodeAPI`, from Kimi Code's docs: k3 and k3-256k are kimi-k3, kimi-for-coding-highspeed is kimi-k2.7-code-highspeed), and kimi-for-coding (K2.8 Preview, not on the API) is unpriced rather than free (#1370). The alias is Kimi Code's alone (`kimiCodePlans`): a relay's own `k3` keeps its own price. A catalog other than these that lists a model at $0 still prices it at $0: models.dev has truly free models (nvidia, modelscope). The Requests list names the API model it was priced as (`PricedNameFor`). | [`models.go`](../../internal/provider/models.go) |
| A plugin's own price | A model no catalog prices (models.dev has no vendor's own model: Devin's SWE-2 and SWE-1.7 Lightning, Cursor's auto, Qoder's efficient) is priced at what its plugin lists it at (`ListPrice` → `pluginListPrice`, `plugin.Model.Cost`, read by `pluginPrice`: an input or output above 0, OpenCode's 0 being none), else at $0 when the plugin says the plan serves it free (`pluginFree`, `plugin.Model.Free`: Devin's Free cost tier, Qoder's and WorkBuddy's x0). Both come after every catalog, so a price a catalog gives is kept over the plugin's. Devin's plugin (0.1.18) lists each paid model at the price Devin's GetCliModelConfigs gives it (input, cached input and output per 1M tokens); a moved Devin user's calls on Devin's own models had no cost on Usage before (太陽と向日葵 on Discord, `TestPluginFreeModelCostsNothing`). A plugin that gives no price and doesn't say free leaves the model unpriced. | [`models.go`](../../internal/provider/models.go), [`plugins.go`](../../internal/provider/plugins.go), [`providers.go`](../../internal/plugin/providers.go) |
| Codex's free auto-review | OpenAI made Codex's Auto-review free for users signed in through a ChatGPT account (silverzhou on Discord), so a `codex-auto-review` call through the Codex subscription is priced at nothing (`ListPrice` → `codexReviewFree`, `provider.ChatGPTReviewFree`), not at GPT-5.6 Luna's price (`pricedNames`), and the Requests list names no model it is priced as (`PricedNameFor`). Not on a Free plan, where it spends the allowance (cybercaius), read from the plan Codex is signed in to now; a Codex signed out since counts as a paid plan, as the gateway's caps take it (`freeOnChatGPT`). The same model through an API key, the ChatGPT API subscription or a relay, and a review that followed the session to another model, keep their price; a price the user set wins. Every call is priced as it is read, so rows logged before are free too; the price views' cache keys on each account's plan (`usagePriceCacheKey`). Guarded by `TestAutoReviewFreeOnChatGPTSignIn` and `usage.TestAutoReviewFreeOnChatGPT`. |
diff --git a/internal/catalog/catalog.go b/internal/catalog/catalog.go
index 9f9883460..6404ecb5b 100644
--- a/internal/catalog/catalog.go
+++ b/internal/catalog/catalog.go
@@ -966,6 +966,113 @@ func canonicallyPriced(all map[string]mdProvider, at int, providers []string, b
return Price{}, false
}
+// madeBy is who a model is, by models.dev's canonical_model_id: the labs
+// that make it, and the providers that list it as their own (ownAPI).
+type madeBy struct{ makers, owners []string }
+
+// makerOwners are, by bare id and by that id with its dots as dashes, the
+// makers of the models models.dev's providers list and those of the
+// providers that are a maker's own API, for the Reset count gen: an entry
+// whose canonical_model_id names the provider listing it, or one it is an
+// API of, as its maker (deepseek's deepseek-flash, deepseek/deepseek-v4.1-flash-1017;
+// meta's muse-spark-1.2; stepfun-ai's step-5-preview), under the entry's
+// id and its canonical's. An entry at $0 counts for neither: a maker's
+// free tier (nvidia's) is not what the model costs.
+var makerOwners struct {
+ sync.Mutex
+ gen int
+ by map[string]*madeBy
+}
+
+// ownAPI is whether the models.dev provider pid is maker's own API: its
+// id, or its id and a region or edition (stepfun-ai is StepFun's global
+// API, alibaba-cn Alibaba's in China). A plan's catalog ("-plan") is not:
+// it lists what a membership charges, not the model's price.
+func ownAPI(pid, maker string) bool {
+ pid, maker = strings.ToLower(pid), strings.ToLower(maker)
+ return (pid == maker || strings.HasPrefix(pid, maker+"-")) && !strings.Contains(pid, "-plan")
+}
+
+// madeByOf is who the bare id b is made and listed as their own by, nil
+// when no provider lists it as its own.
+func madeByOf(all map[string]mdProvider, at int, b string) *madeBy {
+ makerOwners.Lock()
+ defer makerOwners.Unlock()
+ if makerOwners.by == nil || makerOwners.gen != at {
+ by := map[string]*madeBy{}
+ add := func(k, maker, pid string) {
+ for _, k := range []string{k, dashed(k)} {
+ x := by[k]
+ if x == nil {
+ x = &madeBy{}
+ by[k] = x
+ }
+ if !slices.Contains(x.makers, maker) {
+ x.makers = append(x.makers, maker)
+ }
+ if !slices.Contains(x.owners, pid) {
+ x.owners = append(x.owners, pid)
+ }
+ }
+ }
+ for pid, p := range all {
+ for key, m := range p.Models {
+ maker, _, ok := strings.Cut(m.Canonical, "/")
+ if !ok || !ownAPI(pid, maker) || m.Cost == nil || m.Cost.Same(Price{}) {
+ continue
+ }
+ maker = strings.ToLower(maker)
+ add(bareID(key), maker, pid)
+ add(bareID(m.Canonical), maker, pid)
+ }
+ }
+ for _, x := range by {
+ slices.Sort(x.makers)
+ slices.Sort(x.owners)
+ }
+ makerOwners.gen, makerOwners.by = at, by
+ }
+ if x := makerOwners.by[b]; x != nil {
+ return x
+ }
+ return makerOwners.by[dashed(b)]
+}
+
+// MakerPriced is the list price the model's own maker gives it: priced by
+// PricedBy at one provider alone, the first of prefer that is an API of
+// the model's maker (ownAPI), else the first by id of the providers that
+// list it as their own (madeByOf). A model a lab sells itself is costed at
+// the lab's price rather than a reseller's that a maker catalog asked
+// before it lists: Mistral sells DeepSeek's deepseek-v4.1-flash-1017 at
+// $0.30/$1.20, DeepSeek's own price is $0.15/$0.60 (bestony on Discord: a
+// Command Code call costed at the model's official price), and Meta's
+// muse-spark-1.2 and Sakana's fugu-ultra, which no maker preset lists, are
+// priced too. False when no provider lists it as its own, or none of
+// those prices it.
+func MakerPriced(prefer []string, id string) (Price, bool) {
+ all, at := loadGen()
+ b := bareID(id)
+ if r, ok := unprofiled(b); ok {
+ b = r
+ }
+ x := madeByOf(all, at, b)
+ if x == nil {
+ return Price{}, false
+ }
+ var asked []string
+ for _, pid := range prefer {
+ if slices.ContainsFunc(x.makers, func(mk string) bool { return ownAPI(pid, mk) }) {
+ asked = append(asked, pid)
+ }
+ }
+ for _, pid := range append(asked, x.owners...) {
+ if pr, ok := PricedBy([]string{pid}, id); ok {
+ return pr, true
+ }
+ }
+ return Price{}, false
+}
+
// dashed is an id with its dots as dashes: the spelling two ids are compared
// in when one vendor writes a model's version 4.6 and another 4-6.
func dashed(id string) string { return strings.ReplaceAll(id, ".", "-") }
diff --git a/internal/provider/maker_own_price_test.go b/internal/provider/maker_own_price_test.go
new file mode 100644
index 000000000..571a4dcef
--- /dev/null
+++ b/internal/provider/maker_own_price_test.go
@@ -0,0 +1,105 @@
+package provider
+
+import (
+ "os"
+ "path/filepath"
+ "testing"
+
+ "github.com/yetone/magpie/internal/catalog"
+)
+
+// makerOwnCatalog is models.dev's api.json as magpie had it on 2026-10-29,
+// these entries byte for byte. DeepSeek sells V4.1 Flash 1017 as
+// deepseek-flash at $0.15/$0.60 and Mistral resells it, by a dashed id, at
+// $0.30/$1.20; Mistral resells Zhipu's GLM-5.2 too. Alibaba lists Qwen3.7
+// Max with no canonical and Alibaba (China) with one, at a price of its
+// own. StepFun's global API and its China one each list Step 5 Preview,
+// and its Step Plan lists it with no price. Meta, which no maker preset
+// is, lists Muse Spark 1.2 and a contributor edition; NVIDIA lists its own
+// Nemotron at $0.
+const makerOwnCatalog = `
+{"deepseek":{"id":"deepseek","models":{
+"deepseek-flash":{"id":"deepseek-flash","name":"DeepSeek V4.1 Flash 1017","description":"October 17 update of DeepSeek V4.1 Flash with improved reasoning, coding, agentic tasks, and visual understanding","family":"deepseek-flash","attachment":true,"reasoning":true,"reasoning_options":[{"type":"toggle"},{"type":"effort","values":["low","high","max"]}],"tool_call":true,"interleaved":{"field":"reasoning_content"},"structured_output":true,"temperature":true,"knowledge":"2025-05","release_date":"2026-10-17","last_updated":"2026-10-17","modalities":{"input":["text","image"],"output":["text"]},"open_weights":true,"limit":{"context":1048576,"output":393216},"cost":{"input":0.15,"output":0.6,"reasoning":0.6,"cache_read":0.003},"canonical_model_id":"deepseek/deepseek-v4.1-flash-1017"},
+"deepseek-v4-pro":{"id":"deepseek-v4-pro","name":"DeepSeek V4 Pro","description":"DeepSeek V4 Pro snapshot with million-token context and support for thinking and non-thinking modes","family":"deepseek-thinking","attachment":false,"reasoning":true,"reasoning_options":[{"type":"toggle"},{"type":"effort","values":["low","high","max"]}],"tool_call":true,"interleaved":{"field":"reasoning_content"},"structured_output":true,"temperature":true,"release_date":"2026-08-13","last_updated":"2026-08-22","modalities":{"input":["text"],"output":["text"]},"open_weights":true,"limit":{"context":1048576,"output":393216},"cost":{"input":0.66,"output":1.98,"reasoning":1.98,"cache_read":0.022},"canonical_model_id":"deepseek/deepseek-v4-pro-0813"}}},
+"mistral":{"id":"mistral","models":{
+"deepseek-v4-1-flash-1017":{"id":"deepseek-v4-1-flash-1017","name":"DeepSeek V4.1 Flash 1017","description":"October 17 update of DeepSeek V4.1 Flash with improved reasoning, coding, agentic tasks, and visual understanding","family":"deepseek-flash","attachment":false,"reasoning":true,"reasoning_options":[{"type":"effort","values":["none","low","high","max"]}],"tool_call":true,"structured_output":true,"temperature":true,"knowledge":"2025-05","release_date":"2026-10-17","last_updated":"2026-10-17","modalities":{"input":["text"],"output":["text"]},"open_weights":true,"limit":{"context":1048576,"output":128000},"cost":{"input":0.3,"output":1.2,"cache_read":0.007},"canonical_model_id":"deepseek/deepseek-v4.1-flash-1017"},
+"glm-5-2":{"id":"glm-5-2","name":"GLM-5.2","description":"Open flagship GLM for long-horizon coding agents and million-token context work","family":"glm","attachment":false,"reasoning":true,"reasoning_options":[{"type":"effort","values":["none","high","max"]}],"tool_call":true,"structured_output":true,"temperature":true,"release_date":"2026-06-13","last_updated":"2026-06-13","modalities":{"input":["text"],"output":["text"]},"open_weights":true,"limit":{"context":1048576,"output":131072},"status":"deprecated","cost":{"input":1.4,"output":4.4,"cache_read":0.14},"canonical_model_id":"zhipuai/glm-5.2"}}},
+"zhipuai":{"id":"zhipuai","models":{
+"glm-5.2":{"id":"glm-5.2","name":"GLM-5.2","description":"Open flagship GLM for long-horizon coding agents and million-token context work","family":"glm","attachment":false,"reasoning":true,"reasoning_options":[{"type":"toggle"},{"type":"effort","values":["high","max"]}],"tool_call":true,"interleaved":{"field":"reasoning_content"},"structured_output":true,"temperature":true,"release_date":"2026-06-13","last_updated":"2026-06-13","modalities":{"input":["text"],"output":["text"]},"open_weights":true,"limit":{"context":1000000,"output":131072},"cost":{"input":1.19516,"output":4.18306,"cache_read":0.29879,"cache_write":0},"canonical_model_id":"zhipuai/glm-5.2"}}},
+"alibaba":{"id":"alibaba","models":{
+"qwen3.7-max":{"id":"qwen3.7-max","name":"Qwen3.7 Max","description":"Qwen frontier model tuned for agent frameworks, coding assistants, and long tasks","family":"qwen","attachment":false,"reasoning":true,"reasoning_options":[{"type":"toggle"},{"type":"budget_tokens"}],"tool_call":true,"structured_output":true,"temperature":true,"release_date":"2026-05-21","last_updated":"2026-05-21","modalities":{"input":["text"],"output":["text"]},"open_weights":false,"limit":{"context":1000000,"output":131072},"cost":{"input":2.5,"output":7.5,"cache_read":0.5,"cache_write":3.125}}}},
+"alibaba-cn":{"id":"alibaba-cn","models":{
+"qwen3.7-max":{"id":"qwen3.7-max","name":"Qwen3.7 Max","description":"Qwen frontier model tuned for agent frameworks, coding assistants, and long tasks","family":"qwen","attachment":false,"reasoning":true,"reasoning_options":[{"type":"toggle"},{"type":"budget_tokens","max":262144}],"tool_call":true,"structured_output":true,"temperature":true,"release_date":"2026-05-21","last_updated":"2026-05-21","modalities":{"input":["text"],"output":["text"]},"open_weights":false,"limit":{"context":1000000,"output":131072},"cost":{"input":1.65,"output":4.951,"cache_read":0.33,"cache_write":2.0625},"canonical_model_id":"alibaba/qwen3.7-max"}}},
+"stepfun-ai":{"id":"stepfun-ai","models":{
+"step-5-preview":{"id":"step-5-preview","name":"Step 5 Preview","description":"StepFun's next-generation flagship base model for coding and professional knowledge work, with native text, image, and video input and a 1M-token context window","attachment":true,"reasoning":true,"reasoning_options":[{"type":"effort","values":["low","medium","high"]}],"tool_call":true,"interleaved":{"field":"reasoning_content"},"structured_output":true,"temperature":true,"release_date":"2026-09-16","last_updated":"2026-09-20","modalities":{"input":["text","image","video"],"output":["text"]},"open_weights":true,"limit":{"context":1024000,"input":1024000,"output":65536},"cost":{"input":1,"output":2.7,"cache_read":0.1},"canonical_model_id":"stepfun/step-5-preview"}}},
+"stepfun":{"id":"stepfun","models":{
+"step-5-preview":{"id":"step-5-preview","name":"Step 5 Preview","description":"StepFun's next-generation flagship base model for coding and professional knowledge work, with native text, image, and video input and a 1M-token context window","attachment":true,"reasoning":true,"reasoning_options":[{"type":"effort","values":["low","medium","high"]}],"tool_call":true,"interleaved":{"field":"reasoning_content"},"structured_output":true,"temperature":true,"release_date":"2026-09-16","last_updated":"2026-09-20","modalities":{"input":["text","image","video"],"output":["text"]},"open_weights":true,"limit":{"context":1000000,"input":1000000,"output":65536},"cost":{"input":1.04613,"output":2.98895,"cache_read":0.10461},"canonical_model_id":"stepfun/step-5-preview"}}},
+"stepfun-ai-step-plan":{"id":"stepfun-ai-step-plan","models":{
+"step-5-preview":{"id":"step-5-preview","name":"Step 5 Preview","description":"StepFun's next-generation flagship base model for coding and professional knowledge work, with native text, image, and video input and a 1M-token context window","attachment":true,"reasoning":true,"reasoning_options":[{"type":"effort","values":["low","medium","high"]}],"tool_call":true,"interleaved":{"field":"reasoning_content"},"structured_output":true,"temperature":true,"release_date":"2026-09-16","last_updated":"2026-09-20","modalities":{"input":["text","image","video"],"output":["text"]},"open_weights":true,"limit":{"context":1000000,"input":1000000,"output":65536},"canonical_model_id":"stepfun/step-5-preview"}}},
+"meta":{"id":"meta","models":{
+"muse-spark-1.2":{"id":"muse-spark-1.2","name":"Muse Spark 1.2","description":"Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1 with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows.","family":"muse","attachment":true,"reasoning":true,"reasoning_options":[{"type":"effort","values":["minimal","low","medium","high","xhigh"]}],"tool_call":true,"structured_output":true,"temperature":true,"release_date":"2026-08-05","last_updated":"2026-08-05","modalities":{"input":["text","image","video","pdf","audio"],"output":["text"]},"open_weights":false,"limit":{"context":1048576,"output":131072},"cost":{"input":1.25,"output":4.25,"cache_read":0.15},"canonical_model_id":"meta/muse-spark-1.2"},
+"muse-spark-1.2-contributor":{"id":"muse-spark-1.2-contributor","name":"Muse Spark 1.2 Contributor","description":"Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1 with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows.","family":"muse","attachment":true,"reasoning":true,"reasoning_options":[{"type":"effort","values":["minimal","low","medium","high","xhigh"]}],"tool_call":true,"structured_output":true,"temperature":true,"release_date":"2026-08-05","last_updated":"2026-08-05","modalities":{"input":["text","image","video","pdf","audio"],"output":["text"]},"open_weights":false,"limit":{"context":1048576,"output":131072},"cost":{"input":0.1,"output":0.2,"cache_read":0.002},"canonical_model_id":"meta/muse-spark-1.2"}}},
+"nvidia":{"id":"nvidia","models":{
+"nvidia/nemotron-3-ultra-550b-a55b":{"id":"nvidia/nemotron-3-ultra-550b-a55b","name":"Nemotron 3 Ultra 550B A55B","description":"Largest Nemotron 3 model for maximum open-weight reasoning and agent accuracy","family":"nemotron","attachment":false,"reasoning":true,"reasoning_options":[{"type":"effort","values":["none","medium","high"]},{"type":"budget_tokens","min":-1,"max":32768}],"tool_call":true,"structured_output":true,"temperature":true,"knowledge":"2025-09","release_date":"2026-06-04","last_updated":"2026-06-04","modalities":{"input":["text"],"output":["text"]},"open_weights":true,"limit":{"context":1000000,"output":65536},"cost":{"input":0,"output":0},"canonical_model_id":"nvidia/nemotron-3-ultra-550b-a55b"}}}}`
+
+func useMakerOwnCatalog(t *testing.T) {
+ t.Helper()
+ home := t.TempDir()
+ t.Setenv("HOME", home)
+ t.Setenv("USERPROFILE", home)
+ t.Setenv("XDG_CONFIG_HOME", filepath.Join(home, ".config"))
+ t.Setenv("XDG_CACHE_HOME", filepath.Join(home, ".cache"))
+ os.MkdirAll(filepath.Dir(catalog.CachePath()), 0o755)
+ if err := os.WriteFile(catalog.CachePath(), []byte(makerOwnCatalog), 0o644); err != nil {
+ t.Fatal(err)
+ }
+ catalog.Reset()
+ t.Cleanup(catalog.Reset)
+}
+
+// A subscription's call is estimated at the price the model's own maker
+// lists it at, not at a reseller's that a maker preset lists before it
+// (bestony on Discord: Command Code's calls costed at the official price).
+// Command Code serves deepseek/deepseek-v4.1-flash-1017, which Mistral
+// resells at twice DeepSeek's price, and Meta's muse-spark-1.2, which only
+// Meta prices.
+func TestMakerPriceIsTheMakersOwn(t *testing.T) {
+ useMakerOwnCatalog(t)
+
+ cc := Provider{ID: "commandcode-plan", Name: "Command Code"}
+ for _, c := range []struct {
+ id string
+ input, output, cr float64
+ }{
+ // DeepSeek's deepseek-flash, by its canonical, not Mistral's resale
+ {"deepseek/deepseek-v4.1-flash-1017", 0.15, 0.6, 0.003},
+ {"deepseek-v4-1-flash-1017", 0.15, 0.6, 0.003},
+ // Zhipu's own, not Mistral's glm-5-2
+ {"glm-5-2", 1.19516, 4.18306, 0.29879},
+ {"zai-org/GLM-5.2", 1.19516, 4.18306, 0.29879},
+ // the maker preset's own entry, as before, though only
+ // alibaba-cn names the model's maker
+ {"Qwen/Qwen3.7-Max", 2.5, 7.5, 0.5},
+ // StepFun's global API, the preset asked first, not its plan's
+ {"stepfun/Step-5-Preview", 1, 2.7, 0.1},
+ // a lab no maker preset is
+ {"meta/muse-spark-1.2", 1.25, 4.25, 0.15},
+ {"meta/muse-spark-1.2-contributor", 0.1, 0.2, 0.002},
+ // ids the maker lists as given keep their own prices
+ {"deepseek/deepseek-v4-pro", 0.66, 1.98, 0.022},
+ } {
+ for name, price := range map[string]func(string) (catalog.Price, bool){"MakerPrice": MakerPrice, "ListPrice": cc.ListPrice} {
+ pr, ok := price(c.id)
+ if !ok || pr.Input != c.input || pr.Output != c.output || pr.CacheRead != c.cr {
+ t.Errorf("%s(%q) = %+v, %v; want $%v/$%v/$%v", name, c.id, pr, ok, c.input, c.output, c.cr)
+ }
+ }
+ }
+ // a maker's own $0 entry is no price for the model
+ if pr, ok := MakerPrice("nvidia/nemotron-3-ultra-550b-a55b"); ok {
+ t.Errorf("MakerPrice(nemotron) = %+v; want unpriced", pr)
+ }
+ if pr, ok := MakerPrice("thinkingmachines/inkling"); ok {
+ t.Errorf("MakerPrice(inkling) = %+v; want unpriced, no catalog lists it", pr)
+ }
+}
diff --git a/internal/provider/models.go b/internal/provider/models.go
index e23386006..e8d8fd468 100644
--- a/internal/provider/models.go
+++ b/internal/provider/models.go
@@ -1875,7 +1875,7 @@ func (p Provider) ListPrice(model string) (catalog.Price, bool) {
if pr, ok := catalog.PricedBy(priceCatalogs(cats), m); ok {
return pr, true
}
- if pr, ok := catalog.PricedBy(makerPriceCatalogs(), m); ok {
+ if pr, ok := makerListPrice(m); ok {
return pr, true
}
}
@@ -2146,13 +2146,27 @@ var kimiCodeAPI = map[string]string{
// gone since.
func MakerPrice(model string) (catalog.Price, bool) {
for _, m := range pricedNames(model) {
- if pr, ok := catalog.PricedBy(makerPriceCatalogs(), m); ok {
+ if pr, ok := makerListPrice(m); ok {
return pr, true
}
}
return catalog.Price{}, false
}
+// makerListPrice is a model id's list price at its maker: the price of
+// the models.dev provider that lists it as its own model
+// (catalog.MakerPriced), else the first maker preset's that lists it. The
+// lab's own comes first, so a model a maker preset resells (Mistral's
+// DeepSeek V4.1 Flash) is costed at what its lab charges, the official
+// price a subscription's call is estimated at (bestony on Discord, on
+// Command Code).
+func makerListPrice(m string) (catalog.Price, bool) {
+ if pr, ok := catalog.MakerPriced(makerPriceCatalogs(), m); ok {
+ return pr, true
+ }
+ return catalog.PricedBy(makerPriceCatalogs(), m)
+}
+
// SamePriceIn is the maker's list price of the model the user said
// providerID's model is the same as (settings' ModelSameAs, #2344): for an
// id of the provider's own that no list prices, as WorkBuddy's kimi-k3-1
@@ -2357,7 +2371,7 @@ func pricedNames(model string) []string {
func PricedName(model string) string {
n := pricedNames(model)
for _, name := range n {
- if _, ok := catalog.PricedBy(makerPriceCatalogs(), name); ok {
+ if _, ok := makerListPrice(name); ok {
return name
}
}