Skip to content

Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.

Reconstruct Rust terminal reasoning token usage

Magnus Müller7eb26f03395b
diff --git a/RUST_AGENT_INTEGRATION_FEATURES.md b/RUST_AGENT_INTEGRATION_FEATURES.md
index f0926a465..f36193908 100644
--- a/RUST_AGENT_INTEGRATION_FEATURES.md
+++ b/RUST_AGENT_INTEGRATION_FEATURES.md
@@ -709,10 +709,15 @@ Terminal core branch: `magnus/browser-use-rust-main-integration` at terminal mai
- This keeps pre-compaction model outputs, tool actions, tool results, usage, attachments, and final results off Python-visible history surfaces while preserving post-compaction run data.
- Proof: `test_rust_history_applies_terminal_session_compaction_boundary`.
+140. Rust terminal reasoning-token usage reconstruction
+ - Terminal `token_count` usage snapshots now preserve `reasoning_output_tokens` on Browser Use `UsageSummary` surfaces.
+ - The wrapper treats reasoning tokens as completion usage, matching Browser Use's LLM usage accounting, and preserves terminal-reported cumulative `total_tokens` instead of recomputing totals from visible output tokens only.
+ - Proof: `test_rust_history_reconstructs_terminal_reasoning_token_usage`.
+
## Current Verification
- `python3 -m py_compile browser_use/agent/service.py browser_use/rust/service.py browser_use/rust/__init__.py browser_use/__init__.py tests/ci/test_rust_agent.py examples/rust_agent/basic.py examples/rust_agent/real_v8_smoke.py`
-- `uv run pytest -q tests/ci/test_rust_agent.py` (111 tests)
+- `uv run pytest -q tests/ci/test_rust_agent.py` (112 tests)
- `cargo build -q -p browser-use-cli` on terminal branch `magnus/browser-use-rust-main-integration`
- `cargo test -q -p browser-use-cli run_codex_session_command_accepts_task_id_and_model -- --nocapture`
- `CARGO_INCREMENTAL=0 cargo test -q -p browser-use-agent selected_remote_cdp_mode_allows_remote_cdp_connect -- --nocapture`
diff --git a/browser_use/rust/service.py b/browser_use/rust/service.py
index d4cc640e5..257e63107 100644
--- a/browser_use/rust/service.py
+++ b/browser_use/rust/service.py
@@ -1533,10 +1533,32 @@ def _model_usage_payload(payload: dict[str, Any]) -> dict[str, Any]:
return usage if isinstance(usage, dict) else payload
+def _reasoning_output_tokens(usage: dict[str, Any]) -> int:
+ return _int_value(
+ usage.get('reasoning_output_tokens')
+ or usage.get('output_reasoning_tokens')
+ or usage.get('reasoning_tokens')
+ )
+
+
+def _usage_completion_tokens(usage: dict[str, Any]) -> int:
+ return _int_value(usage.get('output_tokens')) + _reasoning_output_tokens(usage)
+
+
+def _usage_total_tokens(usage: dict[str, Any], input_tokens: int, completion_tokens: int) -> int:
+ reported_total = usage.get('total_tokens')
+ if reported_total is not None:
+ total_tokens = _int_value(reported_total)
+ if total_tokens > 0 or input_tokens + completion_tokens == 0:
+ return total_tokens
+ return input_tokens + completion_tokens
+
+
def _usage_from_events(events: list[dict[str, Any]], model: str) -> UsageSummary:
input_tokens = 0
cached_input_tokens = 0
- output_tokens = 0
+ completion_tokens = 0
+ total_tokens = 0
cost = 0.0
invocations = 0
token_count_invocations = 0
@@ -1546,9 +1568,12 @@ def _usage_from_events(events: list[dict[str, Any]], model: str) -> UsageSummary
payload = _event_payload(event)
if event_type == 'model.usage':
usage = _model_usage_payload(payload)
- input_tokens += _int_value(usage.get('input_tokens'))
+ event_input_tokens = _int_value(usage.get('input_tokens'))
+ event_completion_tokens = _usage_completion_tokens(usage)
+ input_tokens += event_input_tokens
cached_input_tokens += _int_value(usage.get('input_cached_tokens') or usage.get('cached_input_tokens'))
- output_tokens += _int_value(usage.get('output_tokens'))
+ completion_tokens += event_completion_tokens
+ total_tokens += _usage_total_tokens(usage, event_input_tokens, event_completion_tokens)
cost += _float_value(usage.get('cost_usd') or usage.get('cost') or payload.get('cost_usd') or payload.get('cost'))
invocations += 1
continue
@@ -1558,17 +1583,16 @@ def _usage_from_events(events: list[dict[str, Any]], model: str) -> UsageSummary
continue
input_tokens = _int_value(token_usage.get('input_tokens'))
cached_input_tokens = _int_value(token_usage.get('cached_input_tokens') or token_usage.get('input_cached_tokens'))
- output_tokens = _int_value(token_usage.get('output_tokens'))
+ completion_tokens = _usage_completion_tokens(token_usage)
+ total_tokens = _usage_total_tokens(token_usage, input_tokens, completion_tokens)
token_count_invocations += 1
invocations = max(invocations, token_count_invocations)
-
- total_tokens = input_tokens + output_tokens
by_model = {
model: ModelUsageStats(
model=model,
prompt_tokens=input_tokens,
- completion_tokens=output_tokens,
+ completion_tokens=completion_tokens,
total_tokens=total_tokens,
cost=cost,
invocations=invocations,
@@ -1579,7 +1603,7 @@ def _usage_from_events(events: list[dict[str, Any]], model: str) -> UsageSummary
total_prompt_cost=0.0,
total_prompt_cached_tokens=cached_input_tokens,
total_prompt_cached_cost=0.0,
- total_completion_tokens=output_tokens,
+ total_completion_tokens=completion_tokens,
total_completion_cost=0.0,
total_tokens=total_tokens,
total_cost=cost,
diff --git a/tests/ci/test_rust_agent.py b/tests/ci/test_rust_agent.py
index b415e0e8d..ccc1f1b07 100644
--- a/tests/ci/test_rust_agent.py
+++ b/tests/ci/test_rust_agent.py
@@ -1032,6 +1032,51 @@ def test_rust_history_reconstructs_terminal_token_count_usage():
assert history.usage.by_model['gpt-test'].completion_tokens == 31
+def test_rust_history_reconstructs_terminal_reasoning_token_usage():
+ from browser_use.rust.service import _history_from_events
+
+ history = _history_from_events(
+ [
+ {
+ 'event_type': 'token_count',
+ 'payload': {
+ 'info': {
+ 'last_token_usage': {
+ 'cached_input_tokens': 2,
+ 'input_tokens': 50,
+ 'output_tokens': 20,
+ 'reasoning_output_tokens': 5,
+ 'total_tokens': 75,
+ },
+ 'total_token_usage': {
+ 'cached_input_tokens': 10,
+ 'input_tokens': 100,
+ 'output_tokens': 40,
+ 'reasoning_output_tokens': 15,
+ 'total_tokens': 155,
+ },
+ }
+ },
+ },
+ {'event_type': 'session.done', 'payload': {'result': 'final answer'}},
+ ],
+ model='gpt-test',
+ started=1.0,
+ finished=2.0,
+ output_model_schema=None,
+ process_error=None,
+ )
+
+ assert history.usage is not None
+ assert history.usage.total_prompt_tokens == 100
+ assert history.usage.total_prompt_cached_tokens == 10
+ assert history.usage.total_completion_tokens == 55
+ assert history.usage.total_tokens == 155
+ assert history.usage.entry_count == 1
+ assert history.usage.by_model['gpt-test'].completion_tokens == 55
+ assert history.usage.by_model['gpt-test'].total_tokens == 155
+
+
def test_rust_history_reconstructs_terminal_nested_model_usage():
from browser_use.rust.service import _history_from_events