Summary
Two new Google Gemini 3.8 Live API models launched in September 2026 are missing from model_list.json. The catalog already tracks gemini-3.8-flash (line ~12096) as well as OpenAI's live model gpt-live-1 (line ~1794, "flavor": "live"), confirming both the model family and the live flavor pattern.
Missing Models
| Model ID |
Type |
Notes |
gemini-3.8-live |
Live API (real-time voice agents) |
Default Live API model, low-latency |
gemini-3.8-live-extended-thinking |
Live API with reasoning |
High-reasoning Live API model |
Suggested Catalog Entries
"gemini-3.8-live": {
"format": "google",
"flavor": "live",
"input_cost_per_mil_tokens": 0.75,
"output_cost_per_mil_tokens": 4.50,
"displayName": "Gemini 3.8 Live",
"available_providers": ["google"]
},
"gemini-3.8-live-extended-thinking": {
"format": "google",
"flavor": "live",
"input_cost_per_mil_tokens": 0.75,
"output_cost_per_mil_tokens": 4.50,
"reasoning": true,
"displayName": "Gemini 3.8 Live Extended Thinking",
"available_providers": ["google"]
}Verification Checklist
| Check |
Status |
Detail |
| Cross-source |
✅ |
Confirmed on Google AI models page (listed as "New Stable", Sep 2026) AND Google AI pricing page (text pricing: $0.75 input / $4.50 output per 1M tokens) |
| Already in catalog |
✅ |
Grepped model_list.json for gemini-3.8-live — no matches. Only gemini-3.8-flash exists. |
| ID format |
✅ |
Follows standard Gemini naming convention (gemini-{version}-{variant}) |
| Not deprecated |
✅ |
Newly launched September 2026, listed as "Stable" |
| Not embedding/CV-only |
✅ |
Live API models for real-time voice/text dialogue |
| Duplicate check |
✅ |
No open or closed issues cover these model IDs |
Pricing Notes
These models have hybrid pricing — token-based for text and per-minute for audio:
- Text: $0.75 input / $4.50 output per 1M tokens
- Audio input: $3.00 per 1M tokens or $0.005/min
- Audio output: $12.00 per 1M tokens or $0.018/min
The suggested entries above include only the text token pricing, consistent with how gpt-live-1 (no pricing) and gpt-realtime-2.1 (text token pricing only) are handled in the catalog.
Fields NOT verified
max_input_tokens / max_output_tokens: Not published on the Google AI models or pricing pages for these specific live models. Should only be added when officially documented.multimodal: Likely true (supports audio + text input), but not explicitly confirmed.- Vertex AI model IDs (
publishers/google/models/gemini-3.8-live): Not verified — may need separate entries if available on Vertex.
Local Files Inspected
packages/proxy/schema/model_list.json — confirmed gemini-3.8-live and gemini-3.8-live-extended-thinking absent; verified gemini-3.8-flash exists at line ~12096; verified gpt-live-1 live model pattern at line ~1794
Summary
Two new Google Gemini 3.8 Live API models launched in September 2026 are missing from
model_list.json. The catalog already tracksgemini-3.8-flash(line ~12096) as well as OpenAI's live modelgpt-live-1(line ~1794,"flavor": "live"), confirming both the model family and the live flavor pattern.Missing Models
gemini-3.8-livegemini-3.8-live-extended-thinkingSuggested Catalog Entries
Verification Checklist
model_list.jsonforgemini-3.8-live— no matches. Onlygemini-3.8-flashexists.gemini-{version}-{variant})Pricing Notes
These models have hybrid pricing — token-based for text and per-minute for audio:
The suggested entries above include only the text token pricing, consistent with how
gpt-live-1(no pricing) andgpt-realtime-2.1(text token pricing only) are handled in the catalog.Fields NOT verified
max_input_tokens/max_output_tokens: Not published on the Google AI models or pricing pages for these specific live models. Should only be added when officially documented.multimodal: Likely true (supports audio + text input), but not explicitly confirmed.publishers/google/models/gemini-3.8-live): Not verified — may need separate entries if available on Vertex.Local Files Inspected
packages/proxy/schema/model_list.json— confirmedgemini-3.8-liveandgemini-3.8-live-extended-thinkingabsent; verifiedgemini-3.8-flashexists at line ~12096; verifiedgpt-live-1live model pattern at line ~1794