Atlas

Compare Models

GPT-6 Astra vs MiniMax M3.1 Flash

GPT-6 Astra leads by 7.4 capability points (≈95% interval +4.8 to +10.0): a clear lead over MiniMax M3.1 Flash · P(higher) ≈ 100%

Wins 33 of 37 shared benchmarks (4 lost)

Capability vs. relative cost
GPT-6 AstraMiniMax M3.1 FlashClosed weightsOpen weights

Each line runs through one model’s priced reasoning settings, cheapest to most expensive; grey points are other models’ priced settings in the same range. Results whose reasoning effort the source didn’t report are left out wherever a model has named effort levels. Capability and relative cost are the published estimates (relative cost is cost per task as a multiple of GPT-5 (high)); hover a setting, or use the arrow keys, to set it beside the other model’s same setting.

GPT-6 AstravsMiniMax M3.1 Flashcapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins33 – 4 · 37 shared

Comparing matching setups only (67 benchmarks excluded) · suites counted once
← GPT-6 AstraMiniMax M3.1 Flash →

GPT-6 AstravsMiniMax M3.1 Flashwins by categoryData Nov 15, 2026Data: Vals AI, Mercor, Scale AI and 18 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is MiniMax M3.1 Flash minus GPT-6 Astra.

ALE-BenchCode31281495−57.9
NoCoT-Bench NCRI 15.2General QA15968.4−49.8
NoCoT-Bench NCKI (kspine v3)General QA13165.9−45.2
Toloka Arena (v3)Agents67.022.0−45.0
RSI Arena Stage 1 human ratingAgents10951002−44.7
TaxCalcBench TY25Finance & Accounting44.00.0−44.0
GBENCH One-shot Coding GScoreGames79.241.1−38.1
Terminal-Bench 4.0Code59.625.3−34.3
DAXBench v2Code96.863.8−33.0
Humanity's Last Exam DiamondGeneral QA60.628.1−32.5
Visa Agentic Commerce BenchmarkAgents64.433.3−31.1
BuseyBench SVGMultimodal7.34.2−31.0
Code Migration v1.1 (Vals Index v2.2 subset)Code78.048.2−29.8
Hebbia Financial Services BenchmarkFinance & Accounting72.347.0−25.3
o11y-bench 1.1 Pass^3Agents82.864.1−18.7
Agents' Last Exam v1.1 (ALE v1.1)Agents42.629.5−13.1
LiveBench 2.0Agents86.173.1−13.0
τ³-Banking (Mercor run, bm25_grep retrieval)Agents46.736.1−10.6
APEX-Agents 1.1Professional Work75.168.1−7.0
GDPxProfessional Work9.64.0−5.6
Legal Research BenchLegal39.434.1−5.3
PostTrainBench Average (v1.1)Indexes44.339.3−5.0
Excel Modeling Benchmark v1.1Finance & Accounting72.767.8−4.8
APEX-Accounting 1.1Finance & Accounting64.760.0−4.7
Agents' Last Exam (ALE)Agents31.627.0−4.6
APEX-1 (Mercor, mean score)Professional Work69.365.0−4.3
Professional Reasoning Benchmark - LegalLegal48.445.1−3.3
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal9.211.9+2.7
Professional Reasoning Benchmark - FinanceFinance & Accounting47.545.6−1.9
Finance Agent v2.1Finance & Accounting60.458.8−1.6
Vals Legal IndexIndexes68.467.1−1.3
Tax Agent BenchFinance & Accounting63.362.7−0.6
HLAB-Vals-1.1Legal7.17.5+0.4
DeepSWE v1.1Code74.173.7−0.4
HiL-Bench (Human-in-Loop Benchmark)Agents65.365.7+0.4
SWE-Bench Pro V2 — FullCode96.996.6−0.3
Vals Finance IndexIndexes55.655.9+0.3

GPT-6 AstravsMiniMax M3.1 Flashnormalised score marginsData Nov 15, 2026Data: Vals AI, Mercor, Scale AI and 18 othersVillage LLM Atlasatlas.village.dev

Win counts use each benchmark’s declared score direction. Represented suite slices and the derived Capability score are excluded from headline totals. Missing or incompatible setups are excluded, never counted as losses.
Loading Atlas data…