Text ArenaLeaderboard
Answer the same prompt as a rival model; real users vote blind on which answer they prefer.
1,494
48% expected win rate against the board leader (1,507)
#6 of 92
high · board 2026-09-02
Source: ArenaGoogleUnited States· released Sep 2026
Gemini 3.8 Flash
Provisional: 6 comparable tests so far. Ranked once it reaches 6 across 3 counting categories.
Google, United States. One of 14 models from this lab on the Index.
Third-party publishedLow confidence (thin coverage)
Snapshot September 8, 2026
OpenCharts Index
68.9
Provisional: not enough comparable tests or counting categories to rank yet · 6 comparable tests · 0 counting categories
Drop any single test and the Index lands between 59.8 and 75.8.
Needs more results (0 of 8 so far)
Distance to today's frontier on the hardest tests. Not a measure of intelligence.
What it's known for
Gemini 3.8 Flash, in 6 results.
Its widest gap is on Code Arena (WebDev), 57.9% behind the best published result. No comparable results yet in knowledge, agentic, multimodal and long context. Between 30% and 35% behind the best published results, on average.
Strongest category
No category counts yet: a category counts once it has more than one comparable test and the model has taken at least half of them. The scores it has are below.
Best single result
Text Arena
Answer the same prompt as a rival model; real users vote blind on which answer they prefer.
1,494
48% expected win rate against the board leader (1,507) · #6 of 92.
Millions of blind votes on real prompts are the market's measure of what people prefer.
Where it trails
Code Arena (WebDev)
Build a web app from the same prompt as a rival model; real users vote blind on the result.
1,567
57.9 points short of parity with the board leader (1,797) · #18 of 86.
People judging finished apps side by side is the most honest measure of front-end quality.
Not measured yet
No comparable result yet in these 4 categories. A blank is a blank, never a zero, and it does not lower the Index.
- Knowledge
- Agentic
- Multimodal
- Long context
Covered, but not averaged into the Index:
- Reasoning1 of 6 tests · needs 3 · mean 56.6
- Coding3 of 7 tests · needs 4 · mean 74.7
- Math1 of 5 tests · needs 3 · mean 48.0
- Human preferencesingle-test category, shown beside the Index · 96.2
See it at work
What Gemini 3.8 Flash was asked to do, and how it did.
Every test it has taken, by category, with the kind of task it faced, who produced the number and where the result landed against the best published one. The examples are original and illustrative, never items from the datasets themselves.
Human preference · mean 96.2 across 1 comparable test · a single-test category, shown beside the Index and never averaged into it
Every result is read against the best published one on its test: percent scores as a share of the best above chance, Arena ratings as the expected win rate against the board leader, open-ended values on a log scale. Sitting a harder exam never lowers a model. Every number links to the publisher that produced it.
Head to head
Gemini 3.8 Flash against whoever you pick.
Choose a rival. Every counted test both have taken appears side by side as a share of the best published result on that test, with the category means above. A win is a higher share; a gap under 2 points is a tie, because one item on a 45-problem exam is 2.2 points and publishers' own standard errors are of that order.
Gemini 3.8 Flash
provisional·Index 68.9C+
0 : 4
wins · 0 ties · 4 shared
GPT-6 Astra
#1·Index 99.0A+
GPT-6 Astra comes out ahead on 4 of the 4 tests both have taken.
Category means
Reasoning
56.699.6Knowledge
—100.0Coding
74.797.7Math
48.099.8Agentic
—98.5Human preference
96.2—
Shared tests · biggest gaps first
- Code Arena (WebDev)Coding−57.91,5671,797
- ProofBenchMath−51.048.0%99.0%
- CritPtReasoning−41.618.3%31.7%
- SciCodeCoding−3.354.4%56.5%
Bars are distances to the best published result on each test (percent scores above chance, Arena ratings as win rate against the leader, open-ended values on a log scale). Raw values as the publishers report them.
Open GPT-6 AstraScores
Every test, every source
The best published run per benchmark, the raw value as the publisher reports it, who produced the number, how close it comes to the best published result on that test, and where the model sits among every model scored on it. Percent scores are read above chance, Arena ratings as a win rate against the board leader, open-ended values on a log scale. A test with too few models to compare against is shown but not counted.
Category means
- Reasoning56.6 · 1/6
- Knowledge—
- Coding74.7 · 3/7
- Math48.0 · 1/5
- Agentic—
- Multimodal—
- Human preferencebeside96.2 · 1/1
- Long context—
Mean · tests taken of the category's comparable tests. A faded category either rests on fewer than half of them or is a single-test category (marked “beside”), which is shown next to the Index and never averaged into it.
CritPtLeaderboardReasoning
18.3%
56.6% of the best (32.3%)
77th percentile of 78
high
Source: Epoch AI — AI Benchmarking HubSciCodeLeaderboardCoding
54.4%
87.7% of the best (62.0%)
79th percentile of 73 · n = 338
medium
Source: Epoch AI — AI Benchmarking HubCursorBenchLeaderboardCoding
69.2%
94.3% of the best (73.4%)
80th percentile of 21
high · reasoning level: High
Source: Epoch AI — AI Benchmarking HubCode Arena (WebDev)LeaderboardCoding
1,567
21% expected win rate against the board leader (1,797)
80th percentile of 86
high · board 2026-09-05
Source: Arena (LMArena) — Leaderboard DatasetProofBenchLeaderboardMath
48.0%
48.0% of the best (100.0%)
67th percentile of 61
Text ArenaLeaderboardHuman preference
1,494
48% expected win rate against the board leader (1,507)
95th percentile of 92
high · board 2026-09-02
Source: Arena (LMArena) — Leaderboard DatasetTop of the leaderboard
Compare with its neighbours.
The best models, ranked here, working inside Theo.
28 of the models on this page run inside Theo today. Theo picks the right one for each step and always shows which engine answered.