Skip to content

Use Claude Code with your ChatGPT, Kimi, Cursor or Grok subscription via a local Anthropic-compatible proxy

Rust

fix(grok): seed stream usage so Claude Code's bar is not zero

Richard Ayotte210c69bc6c6a
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 3ed283b..05d8941 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -3,6 +3,10 @@ title: Changelog
description: Release notes for claude-code-proxy.
---
+## Unreleased
+
+- Grok streams show estimated input usage from the start and exact usage at completion, so Claude Code's status bar no longer stays at zero input tokens.
+
## v0.1.39 (2026-09-10)
- OpenCode Go users can select 14 additional models, including Grok 4.6, GLM 5.3,
diff --git a/docs/src/content/docs/providers/grok.md b/docs/src/content/docs/providers/grok.md
index b01414c..fc29aea 100644
--- a/docs/src/content/docs/providers/grok.md
+++ b/docs/src/content/docs/providers/grok.md
@@ -32,6 +32,8 @@ ANTHROPIC_SMALL_FAST_MODEL=grok-4.6 \
claude --model grok-4.6
```
+Grok 4.5 and Grok 4.6 have a 500,000 token context window. Claude Code treats unknown model IDs as 200,000 tokens, so set `CLAUDE_CODE_MAX_CONTEXT_TOKENS=500000` when using those IDs. Do not append `[1m]`; that is a one-million-token client hint.
+
## Reasoning and tools
The proxy translates Claude messages, function tools, tool results, thinking controls, token usage, and streaming events. Grok reasoning text appears as Claude Code thinking blocks. Grok supports `none`, `low`, `medium`, and `high` effort levels. `xhigh` is forwarded for `grok-4.6`; higher compatibility levels are mapped to the highest supported Grok level for other registered models.
diff --git a/docs/src/content/docs/using/configure-claude-code.md b/docs/src/content/docs/using/configure-claude-code.md
index 1323a94..c3a0d23 100644
--- a/docs/src/content/docs/using/configure-claude-code.md
+++ b/docs/src/content/docs/using/configure-claude-code.md
@@ -37,6 +37,8 @@ sets the ChatGPT context limit to 272K tokens. Set
`CLAUDE_CODE_AUTO_COMPACT_WINDOW=272000` with `gpt-5.6-sol[1m]` so Claude Code
compacts before the upstream limit.
+Grok 4.5 and Grok 4.6 are 500,000 tokens. Set `CLAUDE_CODE_MAX_CONTEXT_TOKENS=500000` and omit `[1m]`.
+
For a provider and model with a different real context limit, choose a safe value or omit the override. `DISABLE_AUTO_COMPACT=1` disables automatic compaction while preserving manual `/compact`, but the session can then hit the upstream limit.
## Persistent Claude Code settings
diff --git a/src/providers/grok/mod.rs b/src/providers/grok/mod.rs
index 63633b7..e166c74 100644
--- a/src/providers/grok/mod.rs
+++ b/src/providers/grok/mod.rs
@@ -112,6 +112,7 @@ impl Provider for GrokProvider {
upstream,
format!("msg_{}", uuid::Uuid::new_v4().simple()),
requested,
+ count_tokens::count_tokens(&translated),
ctx.monitor.clone(),
ctx.req_id.clone(),
ctx.traffic.clone(),
@@ -226,6 +227,7 @@ impl Provider for GrokProvider {
upstream,
format!("msg_{}", uuid::Uuid::new_v4().simple()),
requested,
+ count_tokens::count_tokens(&translated),
ctx.monitor.clone(),
ctx.req_id.clone(),
ctx.traffic.clone(),
@@ -241,6 +243,7 @@ fn stream_response(
response: client::GrokResponse,
message_id: String,
model: String,
+ estimated_input_tokens: u64,
monitor: Option<MonitorHandle>,
req_id: String,
traffic: Option<Arc<crate::traffic::TrafficCapture>>,
@@ -249,6 +252,7 @@ fn stream_response(
response.into_stream(),
message_id,
model,
+ estimated_input_tokens,
monitor,
req_id,
traffic,
@@ -259,6 +263,7 @@ fn stream_body<S>(
upstream: S,
message_id: String,
model: String,
+ estimated_input_tokens: u64,
monitor: Option<MonitorHandle>,
req_id: String,
traffic: Option<Arc<crate::traffic::TrafficCapture>>,
@@ -270,7 +275,8 @@ where
upstream,
decoder: SseDecoder::default(),
reducer: translate::reducer::Reducer::default(),
- translator: StreamTranslator::new(message_id, model),
+ translator: StreamTranslator::new(message_id, model)
+ .with_estimated_input_tokens(estimated_input_tokens),
terminal: false,
error_sent: false,
monitor,
@@ -618,6 +624,7 @@ mod tests {
upstream,
"msg_1".into(),
"grok-4.5".into(),
+ 0,
Some(monitor.clone()),
"req_1".into(),
None,
@@ -643,6 +650,47 @@ mod tests {
assert_eq!(session.output_tokens, 3);
}
+ #[tokio::test]
+ async fn stream_body_seeds_estimated_input_on_message_start() {
+ let upstream = futures_util::stream::iter(vec![Ok(Bytes::from_static(
+ b"data: {\"type\":\"response.output_text.delta\",\"delta\":\"ok\"}\n\ndata: {\"type\":\"response.completed\",\"response\":{\"usage\":{\"input_tokens\":12,\"output_tokens\":3}}}\n\n",
+ ))]);
+ let response = stream_body(
+ upstream,
+ "msg_1".into(),
+ "grok-4.5".into(),
+ 321,
+ None,
+ "req_1".into(),
+ None,
+ );
+ let body = response.into_body().collect().await.unwrap().to_bytes();
+ let events: Vec<serde_json::Value> = crate::anthropic::sse::parse_sse_events(&body)
+ .into_iter()
+ .filter_map(|event| serde_json::from_str(&event.data).ok())
+ .collect();
+ let started = events
+ .iter()
+ .find(|value| value["type"] == "message_start")
+ .unwrap();
+ let finished = events
+ .iter()
+ .find(|value| value["type"] == "message_delta")
+ .unwrap();
+ assert_eq!(
+ started.pointer("/message/usage/input_tokens"),
+ Some(&serde_json::json!(321))
+ );
+ assert_eq!(
+ finished.pointer("/usage/input_tokens"),
+ Some(&serde_json::json!(12))
+ );
+ assert_eq!(
+ finished.pointer("/usage/output_tokens"),
+ Some(&serde_json::json!(3))
+ );
+ }
+
#[tokio::test]
async fn downstream_event_arrives_before_upstream_completion() {
let (tx, rx) = mpsc::channel(2);
@@ -650,6 +698,7 @@ mod tests {
ChannelStream(rx),
"msg_1".into(),
"grok-4.5".into(),
+ 0,
None,
"req_1".into(),
None,
@@ -713,6 +762,7 @@ mod tests {
upstream,
"msg_1".into(),
"grok-4.5".into(),
+ 0,
None,
"req_1".into(),
None,
@@ -733,6 +783,7 @@ mod tests {
ChannelStream(rx),
"msg_1".into(),
"grok-4.5".into(),
+ 0,
None,
"req_1".into(),
Some(traffic),
@@ -782,6 +833,7 @@ mod tests {
upstream,
"msg_1".into(),
"grok-4.5".into(),
+ 0,
None,
"req_1".into(),
Some(traffic),
@@ -820,6 +872,7 @@ mod tests {
upstream,
"msg_1".into(),
"grok-4.5".into(),
+ 0,
None,
"req_1".into(),
Some(traffic),
@@ -847,6 +900,7 @@ mod tests {
futures_util::stream::iter(vec![Ok(Bytes::copy_from_slice(payload))]),
"msg_1".into(),
"grok-4.5".into(),
+ 0,
None,
"req_1".into(),
Some(traffic),
diff --git a/src/providers/grok/translate/stream.rs b/src/providers/grok/translate/stream.rs
index fa2dc3c..ca80b54 100644
--- a/src/providers/grok/translate/stream.rs
+++ b/src/providers/grok/translate/stream.rs
@@ -99,6 +99,9 @@ pub struct StreamTranslator {
started: bool,
finished: bool,
search_blocks: GrokSearchBlocks,
+ // Seeds Claude Code's live usage counter until the provider returns
+ // authoritative usage in the terminal message_delta.
+ estimated_input_tokens: u64,
}
pub struct LiveStreamTranslator {
@@ -109,10 +112,19 @@ pub struct LiveStreamTranslator {
impl LiveStreamTranslator {
pub fn new(message_id: String, model: String) -> Self {
+ Self::with_estimated_input_tokens(message_id, model, 0)
+ }
+
+ pub fn with_estimated_input_tokens(
+ message_id: String,
+ model: String,
+ estimated_input_tokens: u64,
+ ) -> Self {
Self {
decoder: SseDecoder::default(),
reducer: Reducer::default(),
- renderer: StreamTranslator::new(message_id, model),
+ renderer: StreamTranslator::new(message_id, model)
+ .with_estimated_input_tokens(estimated_input_tokens),
}
}
@@ -152,9 +164,15 @@ impl StreamTranslator {
started: false,
finished: false,
search_blocks,
+ estimated_input_tokens: 0,
}
}
+ pub fn with_estimated_input_tokens(mut self, estimated_input_tokens: u64) -> Self {
+ self.estimated_input_tokens = estimated_input_tokens;
+ self
+ }
+
pub fn render(&mut self, events: Vec<ReducerEvent>) -> anyhow::Result<Vec<u8>> {
if self.finished && !events.is_empty() {
anyhow::bail!("event after terminal completion");
@@ -175,7 +193,7 @@ impl StreamTranslator {
emit(
&mut out,
"message_start",
- serde_json::json!({"type":"message_start","message":{"id":self.message_id,"type":"message","role":"assistant","model":self.model,"content":[],"stop_reason":null,"stop_sequence":null,"usage":{"input_tokens":0,"output_tokens":0}}}),
+ serde_json::json!({"type":"message_start","message":{"id":self.message_id,"type":"message","role":"assistant","model":self.model,"content":[],"stop_reason":null,"stop_sequence":null,"usage":{"input_tokens":self.estimated_input_tokens,"output_tokens":0}}}),
);
}
if matches!(event, ReducerEvent::Finish { .. }) {
@@ -360,16 +378,16 @@ fn render(out: &mut Vec<u8>, event: ReducerEvent, search_blocks: GrokSearchBlock
}
ReducerEvent::Finish {
stop_reason,
+ input_tokens,
output_tokens,
web_search_requests,
x_search_requests,
- ..
} => {
let hosted_search_requests = web_search_requests + x_search_requests;
emit(
out,
"message_delta",
- serde_json::json!({"type":"message_delta","delta":{"stop_reason":stop_reason,"stop_sequence":null},"usage":{"output_tokens":output_tokens,"server_tool_use":{"web_search_requests":hosted_search_requests,"x_search_requests":x_search_requests}}}),
+ serde_json::json!({"type":"message_delta","delta":{"stop_reason":stop_reason,"stop_sequence":null},"usage":{"input_tokens":input_tokens,"output_tokens":output_tokens,"server_tool_use":{"web_search_requests":hosted_search_requests,"x_search_requests":x_search_requests}}}),
);
emit(
out,
@@ -532,4 +550,72 @@ mod tests {
.unwrap();
assert!(String::from_utf8(output).unwrap().contains("first"));
}
+
+ #[test]
+ fn estimated_input_is_visible_at_start_and_provider_usage_is_exact_at_finish() {
+ let mut translator = LiveStreamTranslator::with_estimated_input_tokens(
+ "msg_1".into(),
+ "grok-4.5".into(),
+ 321,
+ );
+ let started = translator
+ .push(b"data: {\"type\":\"response.output_text.delta\",\"delta\":\"ok\"}\n\n")
+ .unwrap();
+ let started = crate::anthropic::sse::parse_sse_events(&started)
+ .into_iter()
+ .filter_map(|event| serde_json::from_str::<serde_json::Value>(&event.data).ok())
+ .find(|value| {
+ value.get("type").and_then(serde_json::Value::as_str) == Some("message_start")
+ })
+ .unwrap();
+ assert_eq!(
+ started.pointer("/message/usage/input_tokens"),
+ Some(&serde_json::json!(321))
+ );
+
+ let finished = translator
+ .push(b"data: {\"type\":\"response.completed\",\"response\":{\"usage\":{\"input_tokens\":12,\"output_tokens\":3}}}\n\n")
+ .unwrap();
+ let finished = crate::anthropic::sse::parse_sse_events(&finished)
+ .into_iter()
+ .filter_map(|event| serde_json::from_str::<serde_json::Value>(&event.data).ok())
+ .find(|value| {
+ value.get("type").and_then(serde_json::Value::as_str) == Some("message_delta")
+ })
+ .unwrap();
+ assert_eq!(
+ finished.pointer("/usage/input_tokens"),
+ Some(&serde_json::json!(12))
+ );
+ assert_eq!(
+ finished.pointer("/usage/output_tokens"),
+ Some(&serde_json::json!(3))
+ );
+ }
+
+ #[test]
+ fn provider_input_tokens_are_visible_on_message_delta() {
+ let mut translator = LiveStreamTranslator::new("msg_1".into(), "grok-4.5".into());
+ let _ = translator
+ .push(b"data: {\"type\":\"response.output_text.delta\",\"delta\":\"ok\"}\n\n")
+ .unwrap();
+ let finished = translator
+ .push(b"data: {\"type\":\"response.completed\",\"response\":{\"usage\":{\"input_tokens\":12,\"output_tokens\":3}}}\n\n")
+ .unwrap();
+ let finished = crate::anthropic::sse::parse_sse_events(&finished)
+ .into_iter()
+ .filter_map(|event| serde_json::from_str::<serde_json::Value>(&event.data).ok())
+ .find(|value| {
+ value.get("type").and_then(serde_json::Value::as_str) == Some("message_delta")
+ })
+ .unwrap();
+ assert_eq!(
+ finished.pointer("/usage/input_tokens"),
+ Some(&serde_json::json!(12))
+ );
+ assert_eq!(
+ finished.pointer("/usage/output_tokens"),
+ Some(&serde_json::json!(3))
+ );
+ }
}