Models by vendor · Google
Gemini models on the same tests, re-tested on a fixed schedule
AI Stupid Level gives every Gemini model on the board the same tasks as every other vendor's models: coding every four hours, reasoning and tool use daily, with an hourly canary probe. The scores are ASL's own, from the 24-hour boards, never the vendor's published numbers.
Called through generativelanguage.googleapis.com on our own account, as an application would.
- 3
- Gemini models on the 24-hour boards
- =1st
- best Combined place among the 24 ranked models, held by Gemini 3.8 Flash
- 100%
- of our probes to the Google API answered in the last 7 days
01
How do Gemini models score on the same tests?
3 Gemini models are on the board. The best placed is Gemini 3.8 Flash, joint 1st of the 24 models on the Combined board in the last 24 hours.
As of 26 Oct 2026, 11:37 UTC · 24-hour boards · score, then place in brackets; “=” marks a statistical tie
| Place | Model | Score | Trend | Coding | Reasoning | Tool use |
|---|---|---|---|---|---|---|
| =1 | Gemini 3.8 Flash | 89 ± 2.3 | holding steady | 91 (=11) | 98 (=1) | 77 (=14) |
| =16 | Gemini 3.5 Flash Lite | 86 ± 2.2 | holding steady | 85 (=22) | 96 (=1) | 78 (=14) |
| 24 | Gemini 4 Argon | 65 ± 2.0 | holding steady | 65 (24) |
Places are the leaderboard's own: walking down a board, a model joins the group above it unless that group's leader is ahead by more than the measurement can resolve, so “joint” places are normal. The same boards hold every vendor's models; sign in for the full leaderboard.
Free account: every ranking over 24 hours, 7 days and 30 days, a 3-model watchlist with email alerts, 1,000 routed requests a month and a Data API key. No card.
Already have an account? Log in
02
Which Gemini models are tracked?
3 Google models have a page of their own, each with its place on every board and its history.
- Gemini 3.8 Flash
- Tracked since 5 Sep 2026. List price $1.50 per 1M input tokens and $7.50 per 1M output tokens. Joint 1st of the 24 models on the Combined board in the last 24 hours, scoring 89 ± 2.3.
- Gemini 3.5 Flash Lite
- Tracked since 5 Sep 2026. List price $0.30 per 1M input tokens and $2.50 per 1M output tokens. Joint 16th of the 24 models on the Combined board in the last 24 hours, scoring 86 ± 2.2.
- Gemini 4 Argon
- Tracked since 13 Oct 2026. List price $4 per 1M input tokens and $20 per 1M output tokens. 24th of the 24 models on the Combined board in the last 24 hours, scoring 65 ± 2.0.
Prices are list prices per 1M tokens on generativelanguage.googleapis.com, cache-miss rate, as the leaderboard's price view quotes them. What a correct result costs over the last 30 days is on the best-value AI model; response times are on the fastest AI model.
03
Which Gemini model is best for coding right now?
On the Coding board, re-tested every four hours, the best-placed Gemini model in the last 24 hours is Gemini 3.8 Flash: joint 11th of the 24 models on the Coding board in the last 24 hours, scoring 91 ± 0.9. Next among Gemini models is Gemini 3.5 Flash Lite, joint 22nd of the 24 models on the Coding board in the last 24 hours.
The coding suite gives a model a small working project and a bug report (7 tasks a sweep) and grades the fix by running the project's own tests, including hidden tests the model never sees. Every vendor's models get the same tasks: see the best LLM for coding across all vendors and how coding is scored.
One OpenAI-compatible endpoint, your own provider keys, 1,000 requests a month free · no card
04
Which Gemini model is best for reasoning?
On the Reasoning board, re-tested daily, the best-placed Gemini model in the last 24 hours is Gemini 3.8 Flash: joint 1st of the 22 models on the Reasoning board in the last 24 hours, scoring 98 ± 2.5. Next among Gemini models is Gemini 3.5 Flash Lite, joint 1st of the 22 models on the Reasoning board in the last 24 hours.
The reasoning suite runs 4 long multi-turn working sessions a day, scored on correctness, recovery after a failed step and continuity: whether the model still honours rules stated earlier in the session, checked by running code rather than by a judge model. See the best LLM for reasoning across all vendors and how reasoning is scored.
05
Which Gemini model is best for agents and tool calling?
On the Tool use board, re-tested daily, the best-placed Gemini model in the last 24 hours is Gemini 3.5 Flash Lite: joint 14th of the 22 models on the Tool use board in the last 24 hours, scoring 78 ± 2.5. Next among Gemini models is Gemini 3.8 Flash, joint 14th of the 22 models on the Tool use board in the last 24 hours.
The tool-use suite runs 9 tasks in sandboxed machines, each with tools to call: either the job is done at the end or it is not. It is scored on task completion, tool selection, parameter accuracy, efficiency, error handling, context awareness and safety. See the best AI model for agents across all vendors and how tool use is scored.
06
Is the Google API down?
One short completion on our own key, from one machine in Germany. Checked every 10 minutes: what an application calling generativelanguage.googleapis.com would see.
- Status
- Answering
- Answered
- 100% of checks in the last 24 hours · 100% over the last 7 days
- Median response
- 590 ms to one short completion on our own key
- Last checked
- 26 Oct 2026, 11:40 UTC · checked every 10 minutes
The 7-day history, every episode and the other five providers are on Google API status and the provider status hub. When a provider stops answering, the Smart Router's fallback order moves traffic to the next model on the board.
Fallback order across every vendor on the board · 1,000 requests a month free · no card
07
Is Gemini getting worse?
The Gemini app is not the Gemini API: we probe the API, which is what applications and the Smart Router call. A provider can change a model behind the same API name, and without continuous re-testing that change is invisible to the people relying on it. AI Stupid Level compares each Gemini model with its own past: every suite has its own Page-Hinkley (a CUSUM-family test) change-point detector, which separates a sustained decline from ordinary run-to-run noise, and the hourly canary probe watches for a sudden collapse. On the Combined board over the last 24 hours, of the 3 Gemini models with a current score, 3 are holding steady. Searches for “is Gemini getting worse” and “AI model nerfed” end at the same place: the current answer, model by model, is on Is Gemini getting worse?, and the method is on AI drift detection.
Free alerts: an email when a watched model's coding score falls five points in a week · 3 models on a free account
08
Compare Gemini models
Every pair page puts two models side by side, board by board, on the same tests over the last 24 hours.
Gemini 3.8 Flash, the best-placed Gemini model on the Combined board, against each other vendor's best-placed model:
Other vendors on the same tests: OpenAI models, Anthropic models, DeepSeek models, Moonshot AI (Kimi) models and Z.ai (GLM) models. The compare hub lists every family, and the best AI model ranks them across vendors.
09
How we test Gemini models
Every model on the board gets the same tasks on the same schedule: coding every four hours, reasoning and tool use daily, with an hourly canary probe. Gemini models are called through generativelanguage.googleapis.com on our own account, exactly as an application would call them, so a score describes the model as it is served, not as it was announced.
The coding suite (7 tasks) grades a bug fix by running the project's own tests, including hidden tests the model never sees. The reasoning suite (4 sessions) scores long multi-turn working sessions on correctness, recovery and continuity. The tool-use suite (9 tasks) runs in sandboxed machines, where the job is either done or not. The Combined score weights coding 50%, reasoning 25% and tool use 25%.
The methodology is public — every weight, statistical test and drift constant. The test tasks are private, so no lab can train on them. Read the methodology and how drift detection works.
10
Questions people ask about Gemini models
How many Gemini models does AI Stupid Level benchmark?
3 Google models are on the board: Gemini 3.8 Flash, Gemini 3.5 Flash Lite and Gemini 4 Argon. Each is called through generativelanguage.googleapis.com on our own account and gets the same tasks as every other vendor's models: coding every four hours, reasoning and tool use daily, with an hourly canary probe.
Which Gemini model has the highest Combined score?
Gemini 3.8 Flash: 89 ± 2.3, joint 1st of the 24 models on the Combined board in the last 24 hours. Next among Gemini models is Gemini 3.5 Flash Lite at 86 ± 2.2, joint 16th of the 24 models on the Combined board in the last 24 hours. The Combined score weights coding 50%, reasoning 25% and tool use 25%; "joint" means the models sharing a place are closer than the measurement can separate.
Which Gemini model is best for coding?
Gemini 3.8 Flash: joint 11th of the 24 models on the Coding board in the last 24 hours, scoring 91 ± 0.9. The coding suite is re-tested every four hours and graded by running the project's own tests, including hidden tests the model never sees. Every vendor's models are on the best LLM for coding.
Which Gemini model is the cheapest?
By list price, Gemini 3.5 Flash Lite: $0.30 per 1M input tokens and $2.50 per 1M output tokens, against $4 and $20 for Gemini 4 Argon. List prices are the cache-miss rate on generativelanguage.googleapis.com. What a correct result costs is a different question, answered over the last 30 days on the best-value AI model.
Is the Google API down?
As of 26 Oct 2026, 11:40 UTC the Google API was answering our probe; 100% of checks in the last 24 hours and 100% over the last 7 days answered. We check every 10 minutes: one short completion on our own key, from one machine in Germany. The 7-day history and every episode are on Google API status.
Is Gemini getting worse?
The Gemini app is not the Gemini API: we probe the API, which is what applications and the Smart Router call. Each Gemini model on the board is compared with its own past: every suite has its own Page-Hinkley (a CUSUM-family test) change-point detector, which separates a sustained decline from ordinary run-to-run noise. On the Combined board over the last 24 hours, of the 3 Gemini models with a current score, 3 are holding steady. The current answer, model by model, is on Is Gemini getting worse?.
How often are Gemini models re-tested?
On the same schedule as every other vendor's models: coding every four hours, reasoning and tool use daily, with an hourly canary probe. The methodology is public — every weight, statistical test and drift constant. The test tasks are private, so no lab can train on them. Every weight is on the methodology page.
Watch Gemini models, or let the router choose
Free account: every ranking over 24 hours, 7 days and 30 days, a 3-model watchlist with email alerts, 1,000 routed requests a month and a Data API key. No card.
Free alerts: an email when a watched model's coding score falls five points in a week.
Routes across every vendor on the board, on your own provider keys · 1,000 requests a month free