- input
- $5.00
- cached
- $0.50
- output
- $30.00
- Context
- 1.1Mtokens
- Released
- 6mo ago
Cost/1M tokens
Benchmarks
DeepSWE is the primary coding benchmark where an exact model evaluation exists. Arena leaderboards and other sourced evaluations appear below; missing scores are not zero.
How it ranks
Against its predecessor, similarly priced models and the models ranked around it.
DeepSWE 1.1
Best pass@1 per model · Higher is better
#11 of 28
- 74.1%
- 73.8%
- 73.6%
- 72.7%
- 69.9%
- 69.6%
- 69.0%
- 68.5%
- 67.5%
- 67.2%
- 67.0%GPT-5.5
Arena Text
Best Arena rating per model · Higher is better
Arena #29
- 1511
Muse WonderSimilar price - 1503Claude Opus 4.6Similar price
- 1501Claude Opus 4.7Similar price
- 1490Claude Opus 5Similar price
- 1482GPT-6.1 SolNearby rank
- 1482Claude Opus 4.8Nearby rank
- 1481GPT-5.5
- 1480
MiMo-V2.6-ProNearby rank - 1478Gemini 3.5 FlashNearby rank
- 1475GPT-5.4Predecessor
Level with GPT-5.4 (predecessor, overlapping 95% CIs) · Ahead of 0 of 4 similarly priced models
Arena WebDev
Best Arena rating per model · Higher is better
Arena #65
- 1691Claude Opus 5Similar price
- 1649Muse WonderSimilar price
- 1557Claude Opus 4.7Similar price
- 1555Claude Opus 4.8Similar price
- 1519GPT-5.6 LunaNearby rank
- 1515
Qwen3.7 MaxNearby rank - 1512GPT-5.5
- 1509Kimi K2.6Nearby rank
- 1508GLM 5.1Nearby rank
- 1462GPT-5.4Predecessor
+50 vs GPT-5.4 (predecessor) · Ahead of 0 of 4 similarly priced models
Arena Vision
Best Arena rating per model · Higher is better
Arena #20
- 1300Claude Opus 4.7Similar price
- 1299Claude Opus 4.6Similar price
- 1286Claude Opus 4.8Similar price
- 1285Claude Opus 5Similar price
- 1285GPT-5.5
- 1283GPT-6 AstraNearby rank
- 1282GPT-5.4Predecessor
Level with GPT-5.4 (predecessor, overlapping 95% CIs) · Ahead of 0 of 4 similarly priced models
Arena Search
Best Arena rating per model · Higher is better
Not ranked on Arena Search yet
Arena Document
Best Arena rating per model · Higher is better
Arena #10
- 1497Claude Opus 4.6Similar price
- 1494Claude Opus 4.7Similar price
- 1486Claude Opus 5Similar price
- 1485Claude Opus 4.8Similar price
- 1481GPT-5.5
- 1480Muse Spark 1.1Nearby rank
- 1478Claude Sonnet 4.6Nearby rank
- 1471GPT-5.4Predecessor
Level with GPT-5.4 (predecessor, overlapping 95% CIs) · Ahead of 0 of 4 similarly priced models
Agent Arena
Best IPS score per model · Higher is better
Not ranked on Agent Arena yet
ARC-AGI-2
Best score per model · Higher is better
#13 of 85
- 89.2%Gemini 3.8 FlashNearby rank
- 88.3%Claude Opus 5Similar price
- 85.0%GPT-5.5
- 84.6%Gemini 3.7 FlashNearby rank
- 84.6%GPT-5.5 ProNearby rank
- 75.8%Claude Opus 4.7Similar price
- 74.0%GPT-5.4Predecessor
- 72.1%Claude Opus 4.8Similar price
- 64.6%Claude Opus 4.6Similar price
+11.0 pts vs GPT-5.4 (predecessor) · Ahead of 3 of 4 similarly priced models
FrontierMath-Tier-4-v2-Private
Best score per model · Higher is better
#16 of 72
- 78.0%GPT-5.5 ProNearby rank
- 73.2%Claude Opus 5Similar price
- 72.5%GPT-5.5
- 70.7%GPT-5.6 TerraNearby rank
- 61.0%GPT-5.6 LunaNearby rank
- 56.1%Claude Opus 4.8Similar price
- 49.0%GPT-5.4Predecessor
- 31.7%Claude Opus 4.7Similar price
- 26.8%Claude Opus 4.6Similar price
+23.5 pts vs GPT-5.4 (predecessor) · Ahead of 3 of 4 similarly priced models
FrontierMath-Tiers-1-3-v2-Private
Best score per model · Higher is better
#16 of 87
- 86.0%GPT-5.6 TerraNearby rank
- 85.6%Claude Opus 5Similar price
- 85.3%GPT-5.5
- 82.5%GPT-5.4 ProNearby rank
- 82.1%GPT-5.6 LunaNearby rank
- 80.0%Claude Opus 4.8Similar price
- 78.6%GPT-5.4Predecessor
- 70.2%Claude Opus 4.7Similar price
- 66.0%Claude Opus 4.6Similar price
+6.7 pts vs GPT-5.4 (predecessor) · Ahead of 3 of 4 similarly priced models
GPQA diamond
Best score per model · Higher is better
#38 of 161
- 93.9%Claude Opus 5Similar price
- 93.3%GPT-5.4Predecessor
- 91.0%Claude Opus 4.8Similar price
- 90.9%GLM 5.3Nearby rank
- 90.8%Kimi K2.6Nearby rank
- 90.7%GPT-5.5
- 90.6%Claude Opus 5.5Nearby rank
- 90.5%Claude Opus 4.6Similar price
- 90.2%Claude Opus 4.7Similar price
-2.6 pts vs GPT-5.4 (predecessor) · Ahead of 2 of 4 similarly priced models
OSWorld 2.0
Best score per model · Higher is better
#5 of 9
- 31.4%Claude Opus 5Similar price
- 18.5%Claude Opus 4.8Similar price
- 13.9%Claude Opus 4.7Similar price
- 13.0%GPT-5.5
- 9.3%Claude Sonnet 4.6Similar price
- 4.6%
MiniMax M3Nearby rank
Ahead of 1 of 4 similarly priced models
OTIS Mock AIME 2024-2025
Best score per model · Higher is better
#77 of 158
- 98.9%Claude Opus 5Similar price
- 98.3%Claude Opus 4.8Similar price
- 97.8%Claude Opus 4.7Similar price
- 97.8%GPT-5.4Predecessor
- 94.4%Claude Opus 4.6Similar price
- 85.8%Claude Sonnet 4.6Nearby rank
- 84.7%Gemini 2.5 ProNearby rank
- 84.4%GPT-5.5
- 84.4%o3Nearby rank
- 84.4%Qwen3.5-FlashNearby rank
-13.3 pts vs GPT-5.4 (predecessor) · Ahead of 0 of 4 similarly priced models
SimpleQA Verified
Best score per model · Higher is better
#16 of 87
- 66.2%Gemini 3.5 FlashNearby rank
- 66.2%Gemini 3.6 FlashNearby rank
- 63.0%GPT-5.5
- 60.7%GPT-6 SolNearby rank
- 60.3%Muse Spark 1.2Nearby rank
- 59.9%Claude Opus 5Similar price
- 53.0%Claude Opus 4.8Similar price
- 51.7%Claude Opus 4.7Similar price
- 47.0%Claude Opus 4.6Similar price
- 45.1%GPT-5.4Predecessor
+17.9 pts vs GPT-5.4 (predecessor) · Ahead of 4 of 4 similarly priced models
Terminal Bench
Best score per model · Higher is better
#1 of 49
- 84.7%GPT-5.5
- 81.8%GPT-5.4Predecessor
- 80.2%Claude Opus 4.7Similar price
- 79.8%Claude Opus 4.6Similar price
- 63.1%Claude Opus 4.5Similar price
- 53.4%Claude Sonnet 4.6Similar price
+2.9 pts vs GPT-5.4 (predecessor) · Ahead of 4 of 4 similarly priced models
DeepSWE 1.1
Primary coding-agent benchmark · 113 tasks · Data generated 2026-09-22
- pass@4
- 88.5%
- Average cost / attempt
- $7.23
- Benchmark runs
- 4
Score vs cost by reasoning effort
Average cost per attempt (USD) against pass rate
All published configurations
| Harness / reasoning | pass@1 | pass@4 | Cost / attempt |
|---|---|---|---|
| mini-swe-agent · xhigh | 67.0% | 88.5% | $7.23 |
| mini-swe-agent · high | 64.4% | 90.3% | $5.10 |
| mini-swe-agent · medium | 54.0% | 77.9% | $2.75 |
| mini-swe-agent · low | 27.0% | 47.8% | $1.20 |
pass@1 is the success rate across scored attempts; pass@4 is the share of tasks solved in at least one of four attempts. Costs are reported by DeepSWE for each configuration. These results are not directly comparable to evaluations using different harnesses. Official source JSON.
Reasoning effort
Arena WebDev
Score by reasoning effort · Higher is better
ARC-AGI-2
Score by reasoning effort · Higher is better
GPQA diamond
Score by reasoning effort · Higher is better
OTIS Mock AIME 2024-2025
Score by reasoning effort · Higher is better
All published scores
- Arena TextArena #291481
Level with GPT-5.4 (predecessor, overlapping 95% CIs)
- high148195% CI 1478 to 1485 · 68,359 votes
- default147695% CI 1472 to 1480 · 69,937 votes
- Arena WebDevArena #651512
+50 vs GPT-5.4 (predecessor)
- high1487Codex Harness95% CI 1481 to 1493 · 17,653 votes
- xhigh1512Codex Harness95% CI 1506 to 1518 · 15,381 votes
- default1455Codex Harness95% CI 1449 to 1461 · 15,613 votes
- Arena VisionArena #201285
Level with GPT-5.4 (predecessor, overlapping 95% CIs)
- high128395% CI 1278 to 1289 · 28,348 votes
- default128595% CI 1279 to 1290 · 30,058 votes
- Arena DocumentArena #101481
Level with GPT-5.4 (predecessor, overlapping 95% CIs)
- high148195% CI 1476 to 1487 · 27,324 votes
- default147995% CI 1474 to 1485 · 27,751 votes
- ARC-AGI-2#13 of 8585.0%
+11.0 pts vs GPT-5.4 (predecessor)
- low33.3%
- medium70.4%
- high83.3%
- xhigh85.0%
- FrontierMath-Tier-4-v2-Private#16 of 7272.5%
+23.5 pts vs GPT-5.4 (predecessor)
- xhigh72.5%
Source:Epoch AI evaluation - FrontierMath-Tiers-1-3-v2-Private#16 of 8785.3%
+6.7 pts vs GPT-5.4 (predecessor)
- xhigh85.3%
Source:Epoch AI evaluation - GPQA diamond#38 of 16190.7%
-2.6 pts vs GPT-5.4 (predecessor)
- none77.3%
- low90.7%
Source:Epoch AI evaluation - OSWorld 2.0#5 of 913.0%
- xhigh13.0%
- OTIS Mock AIME 2024-2025#77 of 15884.4%
-13.3 pts vs GPT-5.4 (predecessor)
- none57.8%
- low84.4%
Source:Epoch AI evaluation - SimpleQA Verified#16 of 8763.0%
+17.9 pts vs GPT-5.4 (predecessor)
- xhigh63.0%
Source:Epoch AI evaluation - Terminal Bench#1 of 4984.7%
+2.9 pts vs GPT-5.4 (predecessor)
- unknown84.7%NexAU-AHE
- unknown83.1%Capy
- unknown82.2%Codex CLI
- unknown82.0%Codex
- unknown66.1%clnkr
Benchmark compilation: Epoch AI, Capabilities & benchmarking (CC BY attribution). Scores are converted to percentages and model names matched to the Vertical Tabs catalog. Individual evaluations are credited to their original sources. External results may carry separate licences and evaluation configurations; do not interpret scores from different harnesses as directly comparable.
Arena ratings: Arena (lmarena-ai/leaderboard-dataset), licensed CC BY 4.0. Ratings come from pairwise human votes and are only comparable within one leaderboard; the interval is Arena's 95% confidence interval. Model names are matched to the Vertical Tabs catalog.