opencode-zen-freex 0.2.13 → 0.2.14

package/MAINTAINING.md

@@ -28,13 +28,19 @@ model whose own vendor publishes 1048576/131072. The ids the tier added on
 against a declared 65536, and `glm-5.3-flash-free` is declared 200000/32000
 while its paid twin `glm-5.3-flash` is declared 1000000/131072 in the same
 catalogue — and a 232685-token prompt was served through the free id, which the
-declared window does not allow.
-
-Both numbers are acted on. The window decides when a conversation is compacted
-and how much of it is ever sent; the output cap is sent as `max_tokens` on every
-request, so a reply is cut at it with a finish reason that reads like the
-model's own decision. A window that is too small is never refused, it is just
-quietly never used, which is why this stayed invisible for so long.
+declared window does not allow. `space-bunny-free` under-reports on a third
+axis, its `input`: the catalogue derives that number from its own pair
+(`context − output`, 524288) while prompts of 535,110, 935,110 and 979,550
+tokens were served through the free id, and the axis is what a host that shows
+a window shows — magpie's `context_window` and `max_input_tokens` both follow
+it, so the derivation was on screen as half a window.
+
+All of these numbers are acted on. The window decides when a conversation is
+compacted and how much of it is ever sent; the output cap is sent as
+`max_tokens` on every request, so a reply is cut at it with a finish reason that
+reads like the model's own decision; the `input` axis is the size a
+window-showing host displays. A window that is too small is never refused, it is
+just quietly never used, which is why this stayed invisible for so long.
 
 `limits.mjs` holds what was measured, per axis, with the catalogue value it was
 measured against. Each axis carries `samples` and `how`; the short of it:
@@ -51,6 +57,7 @@ measured against. Each axis carries `samples` and `how`; the short of it:
 | `ll-alpha-free` | output | 262144 | the ceiling the route's own refusal names, against a declared 65536 |
 | `glm-5.3-flash-free` | context | 1000000 | what this route declares for the same upstream model, with a 232685-token prompt served as witness |
 | `glm-5.3-flash-free` | output | 131072 | what this route declares for the same upstream model |
+| `space-bunny-free` | input | 1048576 | the window this route declares for the same id, with prompts of up to nearly twice the declared input served |
 
 A prompt sized just over the window is refused with
 
@@ -74,6 +81,14 @@ figure is a rough one (characters, not tokens) and must not be read as a count.
 A second upstream words the same fact as "This endpoint's maximum context length
 is …" and breaks the request into text input, tool input and output.
 
+**The input**, the third axis: nothing cheap names it — the refusals on this
+route name the window, and this id's own refusal names nothing at all — so it is
+measured by serving. A prompt sized past the declared input is the measurement
+(535,110, 935,110 and 979,550-token prompts each came back with the needle at
+their end echoed, and `usage.prompt_tokens` is the count that says how much was
+served), and the window the catalogue declares for the same id bounds the value
+from above: a request placed over it is refused.
+
 The ceiling, which is cheaper than the window: ask for `max_tokens: 1000000` on a
 short prompt and stop reading as soon as the answer starts. An upstream that
 validates the ceiling refuses and names it — `max_tokens is too large: 1000000.
@@ -104,7 +119,8 @@ own, which is the intended end state.
 that model's window for the rest of the process. The list is rebuilt at every
 start and hourly after, so a route that changes the window it enforces is
 followed without anyone editing this file. Nothing is persisted: a restart falls
-back to what is measured here.
+back to what is measured here. The `input` axis has no live form: the refusals
+this route sends name a window and a completion ceiling, and never an input cap.
 
 **The same shape on a second upstream.** `mimo-v2.6-flash-free` is declared
 200000/32000 here too, and its window is a route figure like the one above — an

package/README.md

@@ -45,7 +45,7 @@ magpie plugin login opencode-zen-free
 
 插件会去发现 Zen 当前的免费模型，包括它们的上下文窗口和思考（reasoning）等级。用 `opencode-zen-free/<模型名>`，例如 `opencode-zen-free/big-pickle`。
 
-免费模型的目录上限有时被 Zen 少报，而少报的那一边没有谁会报错：`deepseek-v4.1-flash-free` 在目录里是 20 万上下文 / 3.2 万输出，而同一上游模型的付费 id 是 100 万 / 38.4 万。上下文窗口太小只会让宿主安静地早点压缩，输出上限太小会把回复截断成看起来像模型自己停下的样子，所以插件对这类模型按实测与路由自述对齐（依据与复测方法见 `MAINTAINING.md` 的 “Limits Zen under-reports”），并在运行中采用上游自报的窗口。
+免费模型的目录上限有时被 Zen 少报，而少报的那一边没有谁会报错：`deepseek-v4.1-flash-free` 在目录里是 20 万上下文 / 3.2 万输出，而同一上游模型的付费 id 是 100 万 / 38.4 万。上下文窗口太小只会让宿主安静地早点压缩，输出上限太小会把回复截断成看起来像模型自己停下的样子；`input` 这一轴同样会被少报——有些宿主显示的「上下文窗口」取的就是它（magpie 如此），而 `space-bunny-free` 的 `input` 是目录自己按 `context − output` 推出的 52.4 万，实测 97.9 万 token 的 prompt 都被完整服务、结尾的暗号原样复述。所以插件对这类模型按实测与路由自述对齐（依据与复测方法见 `MAINTAINING.md` 的 “Limits Zen under-reports”），并在运行中采用上游自报的窗口。
 
 已下线的模型和 SystemOne 模型会被排除：「下线」指的是状态写着 `retired`/`disabled`，或带着已生效的退役日期。目录里标了 `deprecated` 但 Zen 的列表里还在的模型**照常提供**——`deprecated` 是预告，而「还在不在」由列表说了算（实测 2026-10-16：`muse-spark-1.2-contributor-free` 在目录里是 `deprecated`，Zen 仍在服务它，把它当退役就把一个能用的免费模型藏掉了）。这份列表在宿主每次列举模型时重新获取（magpie 在启动时做一次，之后每小时一次）：Zen 不再免费开放的模型会退出列表，新转为免费的模型会加入列表——**不等** OpenCode 的目录或文档写到它：只要 Zen 的列表里有、id 里带 `-free`，就按 provider 自己的协议（OpenAI 兼容 / chat）提供，目录晚一步也不影响。
 

package/limits.mjs

@@ -12,9 +12,17 @@
 // output cap that is too small cuts replies off with a finish reason that looks
 // like the model's own decision.
 //
+// The input axis is the third, and it fails the same way one layer out. A host
+// that plans what to send reads it as the room a prompt has; a host that shows a
+// model's window reads it as the window itself — measured 2026-10-16, magpie's
+// `context_window` and `max_input_tokens` both follow `limit.input` when one is
+// declared. So an input axis that under-reports is not only never used, it is
+// what a user is shown as the model's size.
+//
 // Only evidence moves a value here, and evidence is sorted:
 //
-//   route       the route's own refusal, in its own words
+//   route       the route's own answer about its own limit: the words of a
+//               refusal, or what its own service demonstrates
 //   route-twin  what this route declares for the same upstream model
 //   vendor      the model vendor's published page
 //   witnessed   a generation somebody watched finish
@@ -128,6 +136,19 @@ export const LIMITS = {
       how: "the same paid twin declares 131072 on this axis",
     },
   },
+  // Measured the day a host's window for this id was reported half its size.
+  // Only the input axis moves: the window and the ceiling are what this route
+  // declares, and what was served agrees with both.
+  "space-bunny-free": {
+    input: {
+      value: 1048576,
+      source: "route",
+      declaredAt: 524288,
+      measured: "2026-10-16",
+      samples: "prompts of 535,110, 935,110 and 979,550 tokens served whole, each with the needle at the end of it echoed back and each counted by `usage.prompt_tokens`, so the declared 524288 is already known to be small; a prompt placed over the window was refused 400, in words that name nothing",
+      how: "the catalogue's 524288 is its own `context − output` (1048576 − 524288) read as a cap, and this route does not enforce it: the samples above run from just past the declared cap to nearly twice it, served whole, and the wall is the 1048576-token window this same catalogue declares for this id. A host that shows the input axis shows the window — magpie's `context_window` and `max_input_tokens` are both this field when one is declared — so a derivation left here is a wrong window on screen.",
+    },
+  },
 };
 
 // "This model's maximum context length is 1048576 tokens. However, you
@@ -443,7 +464,7 @@ export function limitOverrides(
   // an object handed in as `table` carries — evidence for a model nobody
   // declared, from a table that never named it.
   const record = table && Object.hasOwn(table, id) ? table[id] : undefined;
-  for (const axis of ["context", "output"]) {
+  for (const axis of ["context", "input", "output"]) {
     // Own property, like the table lookup two lines up: `declared` is handed in
     // by a caller, and a `context` inherited from a custom prototype — or from a
     // polluted `Object.prototype` — would otherwise be read as a declaration the
@@ -451,7 +472,12 @@ export function limitOverrides(
     // check with a number nobody declared.
     const declaredValue = declared && Object.hasOwn(declared, axis) ? declared[axis] : undefined;
     const entry = record?.[axis];
-    const live = (axis === "context" ? observed : observedOutput)?.get(id);
+    // Only the axes a refusal names have a live form — a window and a
+    // completion ceiling. An input axis is answered from the table alone: the
+    // refusals this route sends name the window that bounds it, never an input
+    // cap of its own, so there is nothing to observe into it.
+    const live =
+      (axis === "context" ? observed : axis === "output" ? observedOutput : undefined)?.get(id);
     // A route statement made while this process was running is the same tier as
     // the table and newer than it, so it is taken as it stands. Everything else
     // has to have been measured against the catalogue in front of us now: the

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-zen-freex",
-  "version": "0.2.13",
+  "version": "0.2.14",
   "description": "OpenCode Zen's free models, for hosts that are not OpenCode: the public credential, automatic discovery and native streaming protocols",
   "type": "module",
   "main": "./index.mjs",