diff --git a/CHANGELOG.md b/CHANGELOG.md
index 5c9114e..5acb5f3 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,11 @@
 # Changelog
 
+## 0.5.79-dev (unreleased)
+
+### Features
+
+- **Per-session model badge on overview row** (#153) — `board overview` now annotates each tongxue's row with a compact `[opus→sonnet]` badge when their session has downgraded. Builds on the alert path from PR #221 — the alert block still shows the full picture below; the badge is the instant-glance signal that matches a specific row. Same noise filters apply: cross-provider switches and `<synthetic>` placeholders don't trigger a badge. Stacks on #221.
+
 ## 0.5.78-dev (unreleased)
 
 ### Features
diff --git a/VERSION b/VERSION
index 29bcb8e..d67801e 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-0.5.78-dev
+0.5.79-dev
diff --git a/lib/board_view.py b/lib/board_view.py
index 846d6dc..1011451 100644
--- a/lib/board_view.py
+++ b/lib/board_view.py
@@ -9,7 +9,7 @@
 from lib.common import validate_identity
 from lib.fmt import error, heading, ok, warn
 from lib.tmux_utils import capture_pane, has_session, pane_command
-from lib.token_usage import collect_runtime_alerts, load_budget_defaults
+from lib.token_usage import collect_runtime_alerts, load_budget_defaults, session_model_badges
 
 SHELL_COMMANDS = {"zsh", "bash", "sh", "-zsh", "-bash", ""}
 SPINNER_RE = re.compile(r"^\s*(⠋|⠙|⠹|⠸|⠼|⠴|⠦|⠧|⠇|⠏|●)", re.MULTILINE)
@@ -149,6 +149,9 @@ def cmd_overview(db: BoardDB) -> None:
     print(heading(f"=== {db.env.project_root.name}  {now} ==="))
     print()
 
+    # Compute once so the per-session loop is just a dict lookup.
+    badges = session_model_badges(db.env.project_root)
+
     # ── sessions ──
     for row in db.query("SELECT name, status, last_heartbeat FROM sessions WHERE name != 'all' ORDER BY name"):
         name, task, last_hb = row[0], row[1], row[2]
@@ -161,7 +164,10 @@ def cmd_overview(db: BoardDB) -> None:
         else:
             task = "(no status)"
 
-        line = f"  {status:12s} {name:<10s} {task}"
+        badge = badges.get(name.lower(), "")
+        badge_str = f" {warn(f'[{badge}]')}" if badge else ""
+
+        line = f"  {status:12s} {name:<10s}{badge_str} {task}"
         if ago:
             line += f"  {ago}"
         if inbox:
diff --git a/lib/token_usage.py b/lib/token_usage.py
index 143269d..6ed31c1 100644
--- a/lib/token_usage.py
+++ b/lib/token_usage.py
@@ -190,6 +190,61 @@ def tongxue_token_summary(
     return agg[0] if agg else None
 
 
+def _short_model_label(model: str) -> str:
+    """Pick a one-word label suitable for a row badge: opus/sonnet/haiku/gpt-5.x/mini/other."""
+    lowered = model.lower()
+    if "mini" in lowered:
+        return "mini"
+    if "haiku" in lowered:
+        return "haiku"
+    if "opus" in lowered:
+        return "opus"
+    if "sonnet" in lowered:
+        return "sonnet"
+    if lowered.startswith("gpt-"):
+        # gpt-5.4-mini already handled above; for plain gpt-5.x return the family marker.
+        # Guard the empty-second-segment case ("gpt-" alone, or "gpt--5.4") so the badge
+        # never renders as just an arrow with nothing on one side.
+        parts = lowered.split("-", 2)
+        candidate = parts[1] if len(parts) > 1 else ""
+        return candidate or model[:8]
+    return model[:8]
+
+
+def session_model_badges(
+    project_root: Path,
+    *,
+    recent_hours: float | None = DEFAULT_RECENT_HOURS,
+) -> dict[str, str]:
+    """Return `{tongxue_name_lower: "opus→sonnet"}` for downgraded sessions only.
+
+    Uses the same tier-aware filtering as `model_state_alerts` so cross-provider
+    switches do not produce badges. Names are lowercased to match how the board DB
+    stores session names. Sessions without a downgrade are omitted; consumers can
+    treat a missing key as "all clear".
+    """
+    sessions = _load_project_sessions(project_root, recent_hours=recent_hours)
+    if not sessions:
+        return {}
+    badges: dict[str, str] = {}
+    for s in aggregate_by_name(sessions):
+        models = [m for m in s.get("models", []) if m]
+        if len(models) < 2:
+            continue
+        first, latest = models[0], models[-1]
+        first_tier = _model_tier(first)
+        latest_tier = _model_tier(latest)
+        if first_tier == 0 or latest_tier == 0:
+            continue
+        if latest_tier >= first_tier:
+            continue
+        name = (s.get("name") or "").lower()
+        if not name:
+            continue
+        badges[name] = f"{_short_model_label(first)}→{_short_model_label(latest)}"
+    return badges
+
+
 def _parse_usage_args(args: list[str]) -> dict[str, Any]:
     parsed: dict[str, Any] = {"detail": False, "budget": 0.0, "warn_pct": DEFAULT_BUDGET_WARN_PCT}
     i = 0
diff --git a/package.json b/package.json
index 9519177..0184253 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "claude-nb",
-  "version": "0.5.78-dev",
+  "version": "0.5.79-dev",
   "description": "Multi-agent coordination framework for Claude Code sessions",
   "engines": {
     "node": ">=18"
diff --git a/pyproject.toml b/pyproject.toml
index 12ca73b..ebab236 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "claude-nb"
-version = "0.5.78.dev0"
+version = "0.5.79.dev0"
 description = "Multi-agent coordination framework for Claude Code sessions"
 requires-python = ">=3.11"
 license = "MIT"
diff --git a/tests/test_token_usage.py b/tests/test_token_usage.py
index 27fa1ae..3d9c157 100644
--- a/tests/test_token_usage.py
+++ b/tests/test_token_usage.py
@@ -8,6 +8,7 @@
 
 from lib.token_usage import (
     _project_slug,
+    _short_model_label,
     aggregate_by_name,
     cmd_usage,
     collect_runtime_alerts,
@@ -15,6 +16,7 @@
     load_budget_defaults,
     model_state_alerts,
     parse_session_usage,
+    session_model_badges,
     tongxue_token_summary,
 )
 
@@ -636,3 +638,119 @@ def test_merges_multiple_jsonls(self, tmp_path):
         assert result["input"] == 30
         assert result["output"] == 150
         assert result["messages"] == 2
+
+
+class TestShortModelLabel:
+    def test_opus(self):
+        assert _short_model_label("claude-opus-4-7") == "opus"
+
+    def test_sonnet(self):
+        assert _short_model_label("claude-sonnet-4-7") == "sonnet"
+
+    def test_haiku(self):
+        assert _short_model_label("claude-haiku-4-5-20251001") == "haiku"
+
+    def test_mini_takes_precedence(self):
+        # gpt-5.4-mini should label as "mini" (the downgrade tier signal), not the family.
+        assert _short_model_label("gpt-5.4-mini") == "mini"
+
+    def test_gpt_family(self):
+        assert _short_model_label("gpt-5.4") == "5.4"
+
+    def test_gpt_bare_falls_back_to_truncated(self):
+        """`gpt-` alone has an empty second segment — must not render as empty string."""
+        assert _short_model_label("gpt-") == "gpt-"
+
+    def test_gpt_double_dash_falls_back(self):
+        """`gpt--5.4` splits to ('gpt', '', '5.4') — empty segment must fall back."""
+        # split("-", 2) gives ['gpt', '', '5.4'], parts[1] is "", must fall back.
+        assert _short_model_label("gpt--5.4") == "gpt--5.4"[:8]
+
+
+class TestSessionModelBadges:
+    def test_empty_when_no_jsonls(self, tmp_path):
+        project_dir = tmp_path / "empty"
+        project_dir.mkdir()
+        with patch("lib.token_usage._find_project_dir", return_value=project_dir):
+            assert session_model_badges(tmp_path / "project") == {}
+
+    def test_no_badge_when_all_clear(self, tmp_path):
+        project_dir = tmp_path / "jsonls"
+        project_dir.mkdir()
+        _write_jsonl(
+            project_dir / "s1.jsonl",
+            [
+                {"type": "custom-title", "customTitle": "alice"},
+                _make_assistant_msg(model="claude-opus-4-7"),
+            ],
+        )
+        with patch("lib.token_usage._find_project_dir", return_value=project_dir):
+            assert session_model_badges(tmp_path / "project") == {}
+
+    def test_emits_badge_for_downgrade(self, tmp_path):
+        project_dir = tmp_path / "jsonls"
+        project_dir.mkdir()
+        _write_jsonl(
+            project_dir / "s1.jsonl",
+            [
+                {"type": "custom-title", "customTitle": "alice"},
+                _make_assistant_msg(model="claude-opus-4-7"),
+                _make_assistant_msg(model="claude-sonnet-4-7"),
+            ],
+        )
+        with patch("lib.token_usage._find_project_dir", return_value=project_dir):
+            badges = session_model_badges(tmp_path / "project")
+        assert badges == {"alice": "opus→sonnet"}
+
+    def test_skips_cross_provider_switch(self, tmp_path):
+        """User-initiated cnb model use shouldn't surface as a badge either."""
+        project_dir = tmp_path / "jsonls"
+        project_dir.mkdir()
+        _write_jsonl(
+            project_dir / "s1.jsonl",
+            [
+                {"type": "custom-title", "customTitle": "alice"},
+                _make_assistant_msg(model="claude-opus-4-7"),
+                _make_assistant_msg(model="deepseek-v4-pro"),
+            ],
+        )
+        with patch("lib.token_usage._find_project_dir", return_value=project_dir):
+            assert session_model_badges(tmp_path / "project") == {}
+
+    def test_name_is_lowercased(self, tmp_path):
+        project_dir = tmp_path / "jsonls"
+        project_dir.mkdir()
+        _write_jsonl(
+            project_dir / "s1.jsonl",
+            [
+                {"type": "custom-title", "customTitle": "Alice"},
+                _make_assistant_msg(model="claude-opus-4-7"),
+                _make_assistant_msg(model="claude-haiku-4-5-20251001"),
+            ],
+        )
+        with patch("lib.token_usage._find_project_dir", return_value=project_dir):
+            badges = session_model_badges(tmp_path / "project")
+        assert "alice" in badges
+        assert badges["alice"] == "opus→haiku"
+
+    def test_merges_multiple_jsonls_per_session(self, tmp_path):
+        """Two JSONLs for the same tongxue should aggregate via aggregate_by_name first."""
+        project_dir = tmp_path / "jsonls"
+        project_dir.mkdir()
+        _write_jsonl(
+            project_dir / "s1.jsonl",
+            [
+                {"type": "custom-title", "customTitle": "bob"},
+                _make_assistant_msg(model="claude-opus-4-7"),
+            ],
+        )
+        _write_jsonl(
+            project_dir / "s2.jsonl",
+            [
+                {"type": "custom-title", "customTitle": "bob"},
+                _make_assistant_msg(model="claude-sonnet-4-7"),
+            ],
+        )
+        with patch("lib.token_usage._find_project_dir", return_value=project_dir):
+            badges = session_model_badges(tmp_path / "project")
+        assert badges == {"bob": "opus→sonnet"}