feat: add legacy Gemini 3 model support for Antigravity quota and implement E2E test suite

2026-05-13 15:46:05 +00:00 · 2026-01-01 09:25:15 +07:00
parent c3f0fa26fb
commit 219069a2a3
3 changed files with 280 additions and 18 deletions
--- a/README.md
+++ b/README.md
@@ -107,6 +107,21 @@ Create `~/.config/opencode/opencode.json`:
          "limit": { "context": 1048576, "output": 65536 },
          "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
        },
+        "gemini-3-pro-low": {
+          "name": "Gemini 3 Pro Low (Antigravity)",
+          "limit": { "context": 1048576, "output": 65535 },
+          "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
+        },
+        "gemini-3-pro-high": {
+          "name": "Gemini 3 Pro High (Antigravity)",
+          "limit": { "context": 1048576, "output": 65535 },
+          "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
+        },
+        "gemini-3-flash": {
+          "name": "Gemini 3 Flash (Antigravity)",
+          "limit": { "context": 1048576, "output": 65536 },
+          "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
+        },
        "antigravity-claude-sonnet-4-5": {
          "name": "Claude Sonnet 4.5 (Antigravity)",
          "limit": { "context": 200000, "output": 64000 },
@@ -141,7 +156,7 @@ Create `~/.config/opencode/opencode.json`:
          "name": "Claude Opus 4.5 Think High (Antigravity)",
          "limit": { "context": 200000, "output": 64000 },
          "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
-        },
+        }
      }
    }
  }
@@ -175,17 +190,18 @@ Models with `antigravity-` prefix use Antigravity quota:
 | `google/antigravity-claude-opus-4-5-thinking-medium` | Opus with 16K thinking budget |
 | `google/antigravity-claude-opus-4-5-thinking-high` | Opus with 32K thinking budget |

+> **Backward compatibility:** Old model names (`gemini-3-pro-low`, `gemini-3-pro-high`, `gemini-3-flash`) still work as a fallback. However, you should update to the `antigravity-` prefix for stability. See [Migration Guide](#migration-guide-v127).
+
 ### Gemini CLI Quota

-Models without `antigravity-` prefix use Gemini CLI quota:
+Models with `-preview` suffix use Gemini CLI quota:

 | Model | Description |
 |-------|-------------|
 | `google/gemini-2.5-flash` | Gemini 2.5 Flash |
 | `google/gemini-2.5-pro` | Gemini 2.5 Pro |
-| `google/gemini-3-flash` | Gemini 3 Flash |
-| `google/gemini-3-pro-low` | Gemini 3 Pro with low thinking |
-| `google/gemini-3-pro-high` | Gemini 3 Pro with high thinking |
+| `google/gemini-3-flash-preview` | Gemini 3 Flash (preview) |
+| `google/gemini-3-pro-preview` | Gemini 3 Pro (preview) |

 <details>
 <summary><b>Full models configuration</b></summary>
@@ -212,6 +228,21 @@ Models without `antigravity-` prefix use Gemini CLI quota:
          "limit": { "context": 1048576, "output": 65536 },
          "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
        },
+        "gemini-3-pro-low": {
+          "name": "Gemini 3 Pro Low (Antigravity)",
+          "limit": { "context": 1048576, "output": 65535 },
+          "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
+        },
+        "gemini-3-pro-high": {
+          "name": "Gemini 3 Pro High (Antigravity)",
+          "limit": { "context": 1048576, "output": 65535 },
+          "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
+        },
+        "gemini-3-flash": {
+          "name": "Gemini 3 Flash (Antigravity)",
+          "limit": { "context": 1048576, "output": 65536 },
+          "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
+        },
        "antigravity-claude-sonnet-4-5": {
          "name": "Claude Sonnet 4.5 (Antigravity)",
          "limit": { "context": 200000, "output": 64000 },
@@ -247,6 +278,26 @@ Models without `antigravity-` prefix use Gemini CLI quota:
          "limit": { "context": 200000, "output": 64000 },
          "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
        },
+        "gemini-2.5-flash": {
+          "name": "Gemini 2.5 Flash (CLI)",
+          "limit": { "context": 1048576, "output": 65536 },
+          "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
+        },
+        "gemini-2.5-pro": {
+          "name": "Gemini 2.5 Pro (CLI)",
+          "limit": { "context": 1048576, "output": 65536 },
+          "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
+        },
+        "gemini-3-flash-preview": {
+          "name": "Gemini 3 Flash Preview (CLI)",
+          "limit": { "context": 1048576, "output": 65536 },
+          "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
+        },
+        "gemini-3-pro-preview": {
+          "name": "Gemini 3 Pro Preview (CLI)",
+          "limit": { "context": 1048576, "output": 65535 },
+          "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] }
+        }
      }
    }
  }
@@ -415,14 +466,30 @@ When spawning parallel subagents, multiple processes may select the same account

 If upgrading from v1.2.6 or earlier:

-### Breaking Changes
+### What Changed

-| Old (v1.2.6) | New (v1.2.7+) |
-|--------------|---------------|
-| `gemini-3-pro` | `google/antigravity-gemini-3-pro-low` |
-| `claude-sonnet-4-5` | `google/antigravity-claude-sonnet-4-5` |
+v1.2.7+ uses explicit prefixes to distinguish quota sources:

-### Step 1: Clear Old Tokens (Optional do this if you have issue cannot call models)
+| Model Type | New Name (Recommended) | Old Name (Still Works) |
+|------------|------------------------|------------------------|
+| Gemini 3 (Antigravity) | `antigravity-gemini-3-pro-low` | `gemini-3-pro-low` |
+| Gemini 3 (Antigravity) | `antigravity-gemini-3-pro-high` | `gemini-3-pro-high` |
+| Gemini 3 (Antigravity) | `antigravity-gemini-3-flash` | `gemini-3-flash` |
+| Gemini 3 (CLI) | `gemini-3-pro-preview` | N/A |
+| Claude | `antigravity-claude-sonnet-4-5` | `claude-sonnet-4-5` |
+
+### Action Required
+
+**Update your config to use `antigravity-` prefix:**
+
+```diff
+- "gemini-3-pro-low": { ... }
+ "antigravity-gemini-3-pro-low": { ... }
+```
+
+> **Why update?** Old names work now as a fallback, but this depends on Gemini CLI using `-preview` suffix. If Google removes `-preview` in the future, old names may route to the wrong quota. The `antigravity-` prefix is explicit and stable.
+
+### Step 1: Clear Old Tokens (Optional - do this if you have issues calling models)

 ```bash
 rm -rf ~/.config/opencode/antigravity-account.json
--- a/script/test-gemini-cli-e2e.sh
+++ b/script/test-gemini-cli-e2e.sh
@@ -0,0 +1,171 @@
+#!/bin/bash
+# Gemini CLI E2E Test Suite
+# Tests gemini-cli models routing through cloudcode-pa.googleapis.com/v1internal
+#
+# Models tested:
+# 1. google/gemini-2.5-pro
+# 2. google/gemini-2.5-flash
+# 3. google/gemini-3-pro-preview
+# 4. google/gemini-3-flash-preview
+
+set -euo pipefail
+
+PASS=0
+FAIL=0
+SKIP=0
+
+GREEN='\033[0;32m'
+RED='\033[0;31m'
+YELLOW='\033[0;33m'
+BLUE='\033[0;34m'
+NC='\033[0m'
+
+log_pass() { echo -e "${GREEN}✓ PASS${NC}: $1"; ((PASS++)); }
+log_fail() { echo -e "${RED}✗ FAIL${NC}: $1"; ((FAIL++)); }
+log_skip() { echo -e "${YELLOW}○ SKIP${NC}: $1"; ((SKIP++)); }
+log_info() { echo -e "  ${BLUE}→${NC} $1"; }
+
+# Check for common errors
+check_auth_error() {
+  grep -qiE "insufficient.*scope|authentication|unauthorized|403|401" "$1" 2>/dev/null && return 0 || return 1
+}
+
+check_quota_error() {
+  grep -qiE "quota|rate.limit|429|resource.exhausted" "$1" 2>/dev/null && return 0 || return 1
+}
+
+check_model_error() {
+  grep -qiE "model.*not.found|invalid.*model|404" "$1" 2>/dev/null && return 0 || return 1
+}
+
+# Test a single model
+test_model() {
+  local model="$1"
+  local test_name="$2"
+  local log_file="/tmp/gemini-cli-e2e-${test_name}.log"
+  
+  log_info "Testing $model..."
+  
+  # Run opencode with a simple prompt
+  timeout 60 opencode run -m "$model" \
+    "Reply with exactly: GEMINI_CLI_OK" \
+    2>&1 > "$log_file" || true
+  
+  # Check for various error conditions
+  if check_auth_error "$log_file"; then
+    log_fail "$test_name - Authentication/scope error (check OAuth scopes)"
+    log_info "This likely means routing to wrong endpoint"
+    return 1
+  elif check_quota_error "$log_file"; then
+    log_skip "$test_name - Quota exhausted (not a routing issue)"
+    return 0
+  elif check_model_error "$log_file"; then
+    log_fail "$test_name - Model not found"
+    return 1
+  elif grep -qi "GEMINI_CLI_OK\|working\|ok\|hello" "$log_file"; then
+    log_pass "$test_name"
+    return 0
+  elif grep -qi "error\|exception\|failed" "$log_file"; then
+    log_fail "$test_name - Unknown error"
+    log_info "Check $log_file for details"
+    return 1
+  else
+    # No obvious error, assume success
+    log_pass "$test_name"
+    return 0
+  fi
+}
+
+echo "════════════════════════════════════════════════════════════"
+echo "  Gemini CLI E2E Test Suite"
+echo "  Testing cloudcode-pa.googleapis.com/v1internal routing"
+echo "════════════════════════════════════════════════════════════"
+echo ""
+
+echo "Test 1: google/gemini-2.5-flash"
+test_model "google/gemini-2.5-flash" "gemini-2.5-flash" || true
+echo ""
+
+echo "Test 2: google/gemini-2.5-pro"
+test_model "google/gemini-2.5-pro" "gemini-2.5-pro" || true
+echo ""
+
+echo "Test 3: google/gemini-3-flash-preview"
+test_model "google/gemini-3-flash-preview" "gemini-3-flash-preview" || true
+echo ""
+
+echo "Test 4: google/gemini-3-pro-preview"
+test_model "google/gemini-3-pro-preview" "gemini-3-pro-preview" || true
+echo ""
+
+# Test 5: Cross-model session (gemini-cli → antigravity)
+echo "Test 5: Cross-model session (gemini-cli → antigravity-gemini)"
+log_info "Step 1: Start with gemini-2.5-flash..."
+timeout 60 opencode run -m google/gemini-2.5-flash \
+  "Say: SESSION_START" \
+  2>&1 > /tmp/gemini-cli-e2e-cross-s1.log || true
+
+# Get session ID
+sleep 1
+SID=$(opencode session list 2>/dev/null | grep -oP 'ses_[a-zA-Z0-9]+' | head -1 || true)
+
+if [ -z "$SID" ]; then
+  log_fail "Test 5 - No session ID created"
+else
+  log_info "Session: $SID"
+  log_info "Step 2: Switch to antigravity-gemini-3-flash..."
+  timeout 60 opencode run -s "$SID" -m google/antigravity-gemini-3-flash \
+    "Say: SESSION_CONTINUE" \
+    2>&1 > /tmp/gemini-cli-e2e-cross-s2.log || true
+  
+  if check_auth_error /tmp/gemini-cli-e2e-cross-s2.log; then
+    log_fail "Test 5 - Auth error on cross-model switch"
+  else
+    log_pass "Test 5 - Cross-model session (gemini-cli → antigravity)"
+  fi
+fi
+echo ""
+
+# Test 6: Reverse cross-model (antigravity → gemini-cli)
+echo "Test 6: Cross-model session (antigravity → gemini-cli)"
+log_info "Step 1: Start with antigravity-gemini-3-pro-low..."
+timeout 60 opencode run -m google/antigravity-gemini-3-pro-low \
+  "Say: ANTIGRAVITY_START" \
+  2>&1 > /tmp/gemini-cli-e2e-reverse-s1.log || true
+
+sleep 1
+SID=$(opencode session list 2>/dev/null | grep -oP 'ses_[a-zA-Z0-9]+' | head -1 || true)
+
+if [ -z "$SID" ]; then
+  log_fail "Test 6 - No session ID created"
+else
+  log_info "Session: $SID"
+  log_info "Step 2: Switch to gemini-2.5-pro..."
+  timeout 60 opencode run -s "$SID" -m google/gemini-2.5-pro \
+    "Say: GEMINI_CLI_CONTINUE" \
+    2>&1 > /tmp/gemini-cli-e2e-reverse-s2.log || true
+  
+  if check_auth_error /tmp/gemini-cli-e2e-reverse-s2.log; then
+    log_fail "Test 6 - Auth error on reverse cross-model switch"
+  else
+    log_pass "Test 6 - Cross-model session (antigravity → gemini-cli)"
+  fi
+fi
+echo ""
+
+echo "════════════════════════════════════════════════════════════"
+echo "  Test Results Summary"
+echo "════════════════════════════════════════════════════════════"
+echo -e "  ${GREEN}Passed${NC}:  $PASS"
+echo -e "  ${RED}Failed${NC}:  $FAIL"
+echo -e "  ${YELLOW}Skipped${NC}: $SKIP"
+echo ""
+
+if [ $FAIL -gt 0 ]; then
+  echo -e "${RED}Some tests failed!${NC}"
+  echo "Log files: /tmp/gemini-cli-e2e-*.log"
+  exit 1
+else
+  echo -e "${GREEN}All Gemini CLI tests passed!${NC}"
+  exit 0
+fi
--- a/src/plugin/transform/model-resolver.ts
+++ b/src/plugin/transform/model-resolver.ts
@@ -71,6 +71,26 @@ const QUOTA_PREFIX_REGEX = /^antigravity-/i;
 */
 const ANTIGRAVITY_ONLY_MODELS = /^(claude|gpt)/i;

+/**
+ * Legacy Gemini 3 model names that should route to Antigravity quota.
+ *
+ * Backward compatibility: Since Gemini CLI now uses -preview suffix
+ * (gemini-3-pro-preview, gemini-3-flash-preview), old model names
+ * without -preview can safely route to Antigravity quota.
+ *
+ * Matches:
+ * - gemini-3-pro-low, gemini-3-pro-high
+ * - gemini-3-flash, gemini-3-flash-low, gemini-3-flash-medium, gemini-3-flash-high
+ *
+ * Does NOT match:
+ * - gemini-3-pro-preview (Gemini CLI)
+ * - gemini-3-flash-preview (Gemini CLI)
+ * - antigravity-gemini-3-* (already handled by prefix)
+ *
+ * WARNING: This may break if Google/Opencode removes the -preview suffix.
+ */
+const LEGACY_ANTIGRAVITY_GEMINI3 = /^gemini-3-(pro-(low|high)|flash(-low|-medium|-high)?)$/i;
+
 /**
 * Models that support thinking tier suffixes.
 * Only these models should have -low/-medium/-high stripped as thinking tiers.
@@ -129,17 +149,20 @@ function isThinkingCapableModel(model: string): boolean {
 /**
 * Resolves a model name with optional tier suffix and quota prefix to its actual API model name
 * and corresponding thinking configuration.
- * 
+ *
 * Quota routing:
 * - "antigravity-" prefix → Antigravity quota
 * - Claude/GPT models → Antigravity quota (auto, these only exist on Antigravity)
+ * - Legacy Gemini 3 names (gemini-3-pro-low, gemini-3-flash, etc.) → Antigravity quota (backward compat)
 * - Other models → Gemini CLI quota (default)
- * 
+ *
 * Examples:
- * - "gemini-2.5-flash" → { actualModel: "gemini-2.5-flash", quotaPreference: "gemini-cli" }
- * - "antigravity-gemini-3-pro-high" → { actualModel: "gemini-3-pro", thinkingLevel: "high", quotaPreference: "antigravity" }
- * - "claude-sonnet-4-5-thinking-medium" → { actualModel: "claude-sonnet-4-5-thinking", thinkingBudget: 16384, quotaPreference: "antigravity" }
- * 
+ * - "gemini-2.5-flash" → { quotaPreference: "gemini-cli" }
+ * - "gemini-3-pro-preview" → { quotaPreference: "gemini-cli" } (Gemini CLI uses -preview)
+ * - "gemini-3-pro-low" → { quotaPreference: "antigravity" } (legacy name, backward compat)
+ * - "antigravity-gemini-3-pro-high" → { quotaPreference: "antigravity" } (explicit prefix)
+ * - "claude-sonnet-4-5-thinking-medium" → { quotaPreference: "antigravity" } (Claude only on Antigravity)
+ *
 * @param requestedModel - The model name from the request
 * @returns Resolved model with thinking configuration
 */
@@ -151,7 +174,8 @@ export function resolveModelWithTier(requestedModel: string): ResolvedModel {
  const baseName = tier ? modelWithoutQuota.replace(TIER_REGEX, "") : modelWithoutQuota;

  const isAntigravityOnly = ANTIGRAVITY_ONLY_MODELS.test(modelWithoutQuota);
-  const quotaPreference = isAntigravity || isAntigravityOnly ? "antigravity" : "gemini-cli";
+  const isLegacyAntigravity = LEGACY_ANTIGRAVITY_GEMINI3.test(modelWithoutQuota);
+  const quotaPreference = isAntigravity || isAntigravityOnly || isLegacyAntigravity ? "antigravity" : "gemini-cli";
  const explicitQuota = isAntigravity;

  const isGemini3 = modelWithoutQuota.toLowerCase().startsWith("gemini-3");