Text ArenaLeaderboard
Answer the same prompt as a rival model; real users vote blind on which answer they prefer.
1,491
48% expected win rate against the board leader (1,507)
#9 of 92
high · board 2026-09-02
Source: ArenaGoogleUnited States· released Aug 2026
Gemini 3.7 Flash
Ranked #12, 24.8% behind the frontier on average.
Google, United States. One of 14 models from this lab on the Index.
Third-party publishedMedium confidence (7+ comparable tests, 3+ Index categories)
Available in Theo as Theo Fast
Theo orchestrates it for the steps it does best and always shows which engine answered.
Snapshot September 8, 2026
OpenCharts Index
75.2
Ranked #12 (9 to 16) of 36 · 24.8% behind the frontier on average · 14 comparable tests · 3 counting categories
Drop any single test and the Index lands between 71.9 and 79.5, anywhere from #9 to #16. Neighbours inside that range are ties.
Needs more results (3 of 8 so far)
Averaging 67.6 on the 3 it has; the score opens at 4 of 8.
Distance to today's frontier on the hardest tests. Not a measure of intelligence.
What it's known for
Gemini 3.7 Flash, in 14 results.
Strongest in reasoning (82.2) and coding (76.2). It trails the frontier most in math, 32.7% behind on average. No comparable results yet in agentic, multimodal and long context. Between 20% and 25% behind the best published results, on average. Drop any single test and it would sit anywhere from #9 to #16.
Strongest category
Reasoning
Hard, novel problems: graduate science, abstract puzzles, expert exams.
82.2
- GPQA Diamond94.8% · 99% of best
- ARC-AGI-195.5% · 97% of best
- ARC-AGI-284.6% · 89% of best
- CritPt14.3% · 44% of best
Mean of 4 of the category's 6 comparable tests. Counts toward the Index.
Best single result
GPQA Diamond
Answer a PhD-level multiple-choice question in physics, chemistry or biology that a web search will not settle.
94.8%
98.7% of the best (95.8%), above chance · #2 of 78.
Expert-level science questions are the closest thing to asking a specialist colleague. A model that gets them right can check a technical claim instead of echoing it.
Where it trails
FrontierMath Tier 4
Solve the hardest FrontierMath tier: problems that take expert mathematicians days.
36.6%
62.5% behind the best published result (97.6%) · #17 of 51.
The deepest end of the set. Progress here signals genuinely new capability.
Not measured yet
No comparable result yet in these 3 categories. A blank is a blank, never a zero, and it does not lower the Index.
- Agentic
- Multimodal
- Long context
Covered, but not averaged into the Index:
- Knowledgesingle-test category, shown beside the Index · 91.5
- Human preferencesingle-test category, shown beside the Index · 95.3
See it at work
What Gemini 3.7 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 95.3 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 3.7 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 3.7 Flash
#12·Index 75.2B
1 : 10
wins · 1 tie · 12 shared
GPT-6 Astra
#1·Index 99.0A+
GPT-6 Astra comes out ahead on 10 of the 12 tests both have taken.
Category means
Reasoning
82.299.6Knowledge
91.5100.0Coding
76.297.7Math
67.399.8Agentic
—98.5Human preference
95.3—
Shared tests · biggest gaps first
- −62.536.6%97.6%
- CritPtReasoning−54.014.3%31.7%
- Code Arena (WebDev)Coding−54.01,5871,797
- ProofBenchMath−41.058.0%99.0%
- −23.671.6%93.7%
- FrontierCodeCoding−18.143.6%53.3%
- ARC-AGI-2Reasoning−11.084.6%95.0%
- SimpleQA VerifiedKnowledge−8.569.2%75.6%
- ARC-AGI-1Reasoning−3.095.5%98.5%
- −2.897.2%100.0%
- SciCodeCoding+2.357.9%56.5%
- GPQA DiamondReasoningtie94.8%95.8%
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
- Reasoning82.2 · 4/6
- Knowledgebeside91.5 · 1/1
- Coding76.2 · 4/7
- Math67.3 · 4/5
- Agentic—
- Multimodal—
- Human preferencebeside95.3 · 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.
GPQA DiamondEpoch-runReasoning
94.8%
98.7% of the best (95.8%), above chance
99th percentile of 78 · n = 198
high
Source: Epoch AI — AI Benchmarking HubARC-AGI-2LeaderboardReasoning
84.6%
89.0% of the best (95.0%)
84th percentile of 51 · n = 120
high
Source: Epoch AI — AI Benchmarking HubARC-AGI-1LeaderboardReasoning
95.5%
97.0% of the best (98.5%)
82th percentile of 51 · n = 100
high
Source: Epoch AI — AI Benchmarking HubCritPtLeaderboardReasoning
14.3%
44.2% of the best (32.3%)
65th percentile of 78
high
Source: Epoch AI — AI Benchmarking HubSimpleQA VerifiedEpoch-runKnowledge
69.2%
91.5% of the best (75.6%)
91th percentile of 57 · n = 1,000
high
Source: Epoch AI — AI Benchmarking HubSciCodeLeaderboardCoding
57.9%
93.3% of the best (62.0%)
93th percentile of 73 · n = 338
medium
Source: Epoch AI — AI Benchmarking HubFrontierCodeLeaderboardCoding
43.6%
81.5% of the best (53.5%)
69th percentile of 27
harness: chisel · reasoning effort: medium
Source: Epoch AI — AI Benchmarking HubCursorBenchLeaderboardCoding
61.6%
83.9% of the best (73.4%)
55th percentile of 21
high · reasoning level: High
Source: Epoch AI — AI Benchmarking HubCode Arena (WebDev)LeaderboardCoding
1,587
23% expected win rate against the board leader (1,797)
85th percentile of 86
high · board 2026-09-05
Source: Arena (LMArena) — Leaderboard DatasetMock AIME 2024–2025Epoch-runMath
97.2%
97.2% of the best (100.0%)
81th percentile of 78 · n = 45
high
Source: Epoch AI — AI Benchmarking HubFrontierMath (Tiers 1–3)Epoch-runMath
71.6%
76.4% of the best (93.7%)
75th percentile of 61 · n = 290
high
Source: Epoch AI — AI Benchmarking HubFrontierMath Tier 4Epoch-runMath
36.6%
37.5% of the best (97.6%)
68th percentile of 51 · n = 48
high
Source: Epoch AI — AI Benchmarking HubProofBenchLeaderboardMath
58.0%
58.0% of the best (100.0%)
82th percentile of 61
Text ArenaLeaderboardHuman preference
1,491
48% expected win rate against the board leader (1,507)
91th percentile of 92
high · board 2026-09-02
Source: Arena (LMArena) — Leaderboard DatasetNearby on 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.