feat: add max reasoning effort level (salvage of #4627)

Salvaged from #4627 by @perejaslav. Core Hermes already supports
reasoning_effort=max; this closes the WebUI parity gap (Anchor-A) — adds
max to VALID_REASONING_EFFORTS + dropdown/chip/reasoning-command, keeps
the degrade ladder for capped models.

Co-authored-by: perejaslav <perejaslav@users.noreply.github.com>

api/config.py

@@ -3026,7 +3026,7 @@ def get_effective_default_model(config_data: dict | None = None) -> str:
 # importing from the agent tree (which may not be installed).  Any drift here
 # will show up in the shared test suite since both sides accept the same set.
 # Keep this WebUI-visible set aligned with hermes-agent#29248.
-VALID_REASONING_EFFORTS = ("minimal", "low", "medium", "high", "xhigh")
+VALID_REASONING_EFFORTS = ("minimal", "low", "medium", "high", "xhigh", "max")
 
 
 def parse_reasoning_effort(effort):
@@ -3631,9 +3631,6 @@ def coerce_reasoning_effort_for_model(
         return ""
     if raw == "none":
         return "none"
-    accepts_max_as_xhigh = raw == "max"
-    if accepts_max_as_xhigh:
-        raw = "xhigh"
     if raw not in VALID_REASONING_EFFORTS:
         return ""
     supported = resolve_model_reasoning_efforts(
@@ -3649,17 +3646,15 @@ def coerce_reasoning_effort_for_model(
     # those paths return a NON-empty clamped set, so the degrade ladder below
     # still applies. When the set is empty we can't tell "unsupported" from
     # "unknown", so preserve the user's configured effort verbatim where it is
-    # still valid. A stale 'max' value is no longer parser-valid on the WebUI
-    # side, so degrade that unknown-model case to xhigh instead of silently
-    # dropping reasoning later in parse_reasoning_effort(). (#3505 review)
+    # still valid. (#3505 review)
     if not supported:
-        return "xhigh" if accepts_max_as_xhigh else raw
+        return raw
     if raw in supported:
-        return "xhigh" if accepts_max_as_xhigh else raw
+        return raw
     # Degrade to the closest *lower* supported level instead of silently
     # disabling reasoning. e.g. max -> xhigh -> high, or xhigh -> high when the
     # target model caps below the configured effort. Never escalate.
-    ladder = list(VALID_REASONING_EFFORTS)  # ascending: minimal..xhigh
+    ladder = list(VALID_REASONING_EFFORTS)  # ascending: minimal..xhigh..max
     try:
         raw_idx = ladder.index(raw)
     except ValueError:
@@ -3712,10 +3707,8 @@ def get_reasoning_status(
     return {
         # Match CLI default (True if unset in config.yaml)
         "show_reasoning": bool(show_raw) if isinstance(show_raw, bool) else True,
-        # Report the COERCED effort (not the raw config value) so boot/status/chip
-        # read paths agree with what streaming actually sends — e.g. a stale
-        # `reasoning_effort: max` surfaces as `xhigh`, not the now-unsupported `max`.
-        # (Codex review of the drop-max alignment.)
+        # Report the COERCED effort so boot/status/chip read paths agree with
+        # what streaming actually sends. (Codex review of the drop-max alignment.)
         "reasoning_effort": coerce_reasoning_effort_for_model(
             str(effort_raw or "").strip().lower(),
             resolve_model,
@@ -3834,7 +3827,7 @@ def set_reasoning_effort(
     """Persist ``agent.reasoning_effort`` to the active profile's config.yaml.
 
     Mirrors CLI ``/reasoning <level>``: same key, same valid values
-    (``none`` | ``minimal`` | ``low`` | ``medium`` | ``high`` | ``xhigh``).
+    (``none`` | ``minimal`` | ``low`` | ``medium`` | ``high`` | ``xhigh`` | ``max``).
     Raises ``ValueError`` on an unrecognised level so callers can return 400.
     """
     raw = str(effort or "").strip().lower()

static/commands.js

@@ -30,7 +30,7 @@ const COMMANDS=[
   {name:'background',desc:t('cmd_background'),fn:cmdBackground,arg:'prompt',  noEcho:true},
   {name:'status',    desc:t('cmd_status'),   fn:cmdStatus},
   {name:'voice',     desc:t('cmd_voice'),    fn:cmdVoice,     noEcho:true},
-  {name:'reasoning', desc:t('cmd_reasoning'), fn:cmdReasoning, arg:'show|hide|none|minimal|low|medium|high|xhigh', subArgs:['show','hide','none','minimal','low','medium','high','xhigh'], noEcho:true},
+  {name:'reasoning', desc:t('cmd_reasoning'), fn:cmdReasoning, arg:'show|hide|none|minimal|low|medium|high|xhigh|max', subArgs:['show','hide','none','minimal','low','medium','high','xhigh','max'], noEcho:true},
   {name:'yolo', desc:t('cmd_yolo'), fn:cmdYolo, noEcho:true},
   {name:'branch', desc:t('cmd_branch'), fn:cmdBranch, arg:'[name]', noEcho:true},
 ];
@@ -1799,13 +1799,13 @@ function cmdReasoning(args){
   const BRAIN='\uD83E\uDDE0';
   // Matches hermes_constants.VALID_REASONING_EFFORTS + 'none' (CLI parity).
   // Keep this WebUI effort list in sync with hermes-agent#29248.
-  const EFFORTS=['none','minimal','low','medium','high','xhigh'];
+  const EFFORTS=['none','minimal','low','medium','high','xhigh','max'];
   // Shared status renderer used by the no-args branch and as a fallback.
   function _fmtStatus(st){
     const vis=(st && st.show_reasoning===false)?'off':'on';
     const eff=(st && st.reasoning_effort)||'default';
     return BRAIN+' Reasoning effort: '+eff+' \u00B7 display: '+vis
-      +'  |  /reasoning show|hide|none|minimal|low|medium|high|xhigh';
+      +'  |  /reasoning show|hide|none|minimal|low|medium|high|xhigh|max';
   }
   if(!arg){
     // Status — read from the same config.yaml keys the CLI uses.

static/index.html

@@ -759,6 +759,7 @@
             <div class="reasoning-option" data-effort="medium">Medium</div>
             <div class="reasoning-option" data-effort="high">High</div>
             <div class="reasoning-option" data-effort="xhigh">Extra High</div>
+            <div class="reasoning-option" data-effort="max">Max</div>
           </div>
           <div class="composer-toolsets-dropdown" id="composerToolsetsDropdown">
             <div class="toolsets-dropdown-desc" id="toolsetsDropdownDesc"></div>

static/ui.js

@@ -4481,6 +4481,7 @@ function _formatReasoningEffortLabel(effort){
   if(effort==='medium') return 'Medium';
   if(effort==='high') return 'High';
   if(effort==='xhigh') return 'XHigh';
+  if(effort==='max') return 'Max';
   return effort.charAt(0).toUpperCase()+effort.slice(1);
 }
 

tests/test_issue1103_reasoning_chip_visibility.py

@@ -45,8 +45,8 @@ def test_reasoning_chip_html_starts_hidden():
         src
     )
     assert m, "composerReasoningWrap must start with style='display:none'"
-    assert 'data-effort="max"' not in src, (
-        "composer reasoning dropdown must not include Max"
+    assert 'data-effort="max"' in src, (
+        "composer reasoning dropdown must include Max option"
     )
 
 

tests/test_reasoning_effort_model_capabilities.py

@@ -83,7 +83,7 @@ def test_coerce_preserves_effort_for_unrecognized_model():
         "max",
         "brand-new-model-2099",
         provider_id="some-custom-provider",
-    ) == "xhigh"
+    ) == "max"
     # 'none' / unset still pass through unchanged for unknown models.
     assert cfg.coerce_reasoning_effort_for_model(
         "none", "some-unknown-model-xyz", provider_id="custom"

tests/test_reasoning_show_hide.py

@@ -312,14 +312,11 @@ class TestReasoningCommand:
         m = re.search(r'function cmdReasoning\(.*?\n\}', src, re.DOTALL)
         assert m
         fn = m.group(0)
-        for level in ('none', 'minimal', 'low', 'medium', 'high', 'xhigh'):
+        for level in ('none', 'minimal', 'low', 'medium', 'high', 'xhigh', 'max'):
             assert f"'{level}'" in fn, (
                 f"cmdReasoning must accept '{level}' (CLI parity with "
                 f"hermes_constants.parse_reasoning_effort)"
             )
-        assert "'max'" not in fn, (
-            "cmdReasoning should only expose efforts up to xhigh"
-        )
 
     def test_reasoning_subargs_match_cli_levels(self):
         """Autocomplete subArgs must expose every CLI effort level + show/hide."""
@@ -328,14 +325,11 @@ class TestReasoningCommand:
         assert m, "reasoning COMMANDS entry not found"
         entry = m.group(0)
         for suggestion in (
-            'show', 'hide', 'none', 'minimal', 'low', 'medium', 'high', 'xhigh'
+            'show', 'hide', 'none', 'minimal', 'low', 'medium', 'high', 'xhigh', 'max'
         ):
             assert f"'{suggestion}'" in entry, (
                 f"reasoning subArgs must include '{suggestion}' for CLI parity"
             )
-        assert "'max'" not in entry, (
-            "reasoning command subArgs must not include 'max'"
-        )
 
 
 # ── api/config.py — reasoning helpers ────────────────────────────────────────
@@ -365,7 +359,7 @@ class TestReasoningConfigHelpers:
         # Snapshot-style assertion: if hermes_constants adds a level, this
         # test will fail fast so we know to update WebUI too.
         assert VALID_REASONING_EFFORTS == (
-            'minimal', 'low', 'medium', 'high', 'xhigh'
+            'minimal', 'low', 'medium', 'high', 'xhigh', 'max'
         )
 
     def test_set_reasoning_effort_persists_to_config_yaml(self, tmp_path, monkeypatch):