Across 27 comparable benchmarks, GPT-6 Astra has better reported results on 23 and worse on 4. 2 further shared results are listed below, each with the reason it is not counted.
Moderate evidence27 of 29 shared results countedHow we compare
Price basis: Listed rates vs standard rates. The price section below shows the applicable periods.
| Fact | GPT-6 Astra | Mistral Large 4 |
|---|---|---|
| Context window | 1.05M tokens | 1M tokens |
| Input / 1M | $10.00 | $0.700 |
| Output / 1M | $50.00 | $2.10 |
| Blended (4:1) | $11.52 | $0.526 |
| Status | Preview | Preview |
| Released | September 3, 2026 | October 6, 2026 |
Blended is 4:1 input to output with 90% cache hits at listed rates β set your own mix.
Mean of 18 shared benchmarks grouped by capability on a 0β100 scale. How averages are built
GPT-6 Astra ahead on 8 of 10 counted
GPT-6 Astra ahead on 3 of 3 counted
GPT-6 Astra ahead on 4 of 4 counted
GPT-6 Astra ahead on 1 of 1 counted
GPT-6 Astra ahead on 3 of 5 counted
No shared comparable results for Abstract reasoning, Computer & tool use, Research & search, Long context, Cybersecurity.
Published scores, with matching modes where available. Gaps show GPT-6 Astra relative to Mistral Large 4. Reading the results
| Benchmark | GPT-6 Astra | Mistral Large 4 | Gap |
|---|---|---|---|
| AA Cyber IndexRepository Engineering Β· Composite index (0β100) | 33.4Max Β· tools | 49.5β Thinking Β· tools | 0.68Γ#15 vs #7 of 22 |
| Code MigrationMaintenance & Optimization Β· Hidden-test pass rate (%) | 67.7Max Β· tools | 30.6β Thinking High Β· tools | +37.2 pts#4 vs #43 of 79 |
| DeepSWERepository Engineering Β· Pass@1 (DeepSWE v1.1) | 74.1Extra-High Β· tools | 61.7β Thinking Β· tools | +12.4 pts#13 vs #45 of 73 |
| FrontierCode 1.1Code Generation & Editing Β· Weighted rubric score (%)Not counted Β· Different or unrecorded test environments β excluded from the tally | 64.5Max Β· toolsΒ· frontiercode-1-1-extended | 39.4β Thinking High Β· toolsΒ· frontiercode-1-1-main | β#2 vs #30 of 51 |
| FrontierCode 2.0Repository Engineering Β· Weighted rubric score | 44.0Max Β· tools + web | 8.8β Thinking High Β· tools + web | 4.99Γ#6 vs #30 of 36 |
| IOI (Vals v2)Algorithmic Coding Β· Subtask points (%) | 100Max Β· tools | 45.3β Thinking High Β· tools | +54.7 pts#1 vs #32 of 45 |
| SWE-Atlas-QnARepository Engineering Β· Task Resolve Rate (%) | 59.1Extra-High Β· tools | 59.4β Thinking Β· tools | β0.3 pts#6 vs #4 of 17 |
| Terminal-Bench 4.0Agentic Development Β· Resolution rate (%)Not counted Β· Different or unrecorded test environments β excluded from the tally | 59.6Unspecified Reasoning Β· toolsΒ· vals-mini-swe | 28.3β Thinking Β· toolsΒ· legacy-default | β#11 vs #40 of 90 |
| Terminal-Bench 4.1Agentic Development Β· Resolution rate (%) | 60.3Max Β· tools | 27.6Max Β· tools | +32.7 pts#8 vs #26 of 48 |
| Vibe Code Bench 1-100Agentic Development Β· Accuracy (%) | 27.6Max Β· tools | 14.3β Thinking High Β· tools | +13.4 pts#5 vs #20 of 25 |
| Vibe Code Bench v1.1Code Generation & Editing Β· Pass rate (%) | 89.6Max Β· tools | 78.4β Thinking High Β· tools | +11.2 pts#10 vs #36 of 121 |
| WeirdML v3ML Engineering Β· Official score (0-1; higher is better) | 0.42Extra-High Β· tools | 0.044β Thinking High Β· tools | 9.50Γ#1 vs #25 of 26 |
| FrontierMath Tier 4 v2Mathematics Β· Accuracy (verification_code) | 97.6Thinking High Β· tools | 26.8Thinking High Β· tools | +70.8 pts#4 vs #40 of 73 |
| FrontierMath v2Mathematics Β· Accuracy (verification_code) | 93.7Max Β· tools | 58.6β Thinking High Β· tools | +35.1 pts#2 vs #46 of 85 |
| ProofBench v1.1 (Lean 4)Mathematics Β· Accuracy (%) | 99.0Max Β· tools | 10.0β Thinking High Β· tools | +89.0 pts#6 vs #49 of 54 |
| MysteryMechanismScientific Reasoning Β· Accuracy (%) | 53.1Max Β· tools | 12.2β Thinking High Β· tools | +41.0 pts#2 vs #32 of 32 |
| BioMysteryBenchBiology & Genomics Β· Accuracy (%) | 79.3Max Β· tools | 67.0β Thinking High Β· tools | +12.2 pts#8 vs #21 of 28 |
| MedCodeClinical Workflows Β· Accuracy (%) | 48.5Max Β· tools | 40.7β Thinking High Β· no tools | +7.8 pts#29 vs #58 of 97 |
| MedScribeClinical Workflows Β· Rubric pass rate (%) | 87.9Max Β· tools | 80.4β Thinking High Β· no tools | +7.5 pts#16 vs #53 of 99 |
| MMMU-ProVisual Understanding Β· Accuracy | 87.8Max Β· tools | 76.4β Thinking Β· no tools | +11.4 pts#5 vs #55 of 136 |
| EMB (Excel Modeling)Finance Β· Accuracy (%) | 71.7Max Β· tools | 56.2β Thinking High Β· tools | +15.5 pts#13 vs #49 of 78 |
| Finance Agent v2Finance Β· Score | 53.5Max Β· tools | 54.7β Thinking High Β· tools | 0.98Γ#32 vs #24 of 75 |
| Harvey's Legal Agent BenchmarkLegal Β· Accuracy (%) | 5.4Max Β· tools | 15.8β Thinking High Β· tools | β10.4 pts#26 vs #6 of 52 |
| Legal Research BenchLegal Β· Score | 39.4Max Β· tools | 31.7β Thinking High Β· tools | 1.24Γ#38 vs #52 of 91 |
| Tax Agent BenchFinance Β· Accuracy (%, rubric Γ citation quality) | 63.3Max Β· tools | 63.3β Thinking High Β· tools | +0.1 pts#13 vs #14 of 74 |
| AA Intelligence Index v5.0.1Capability Indices Β· Index (0-100) | 56.9Max Β· task tools | 25.3β Thinking Β· tools | 2.24Γ#7 vs #39 of 62 |
| LiveBenchCross-capability Suites Β· Accuracy | 82.2Max Β· no tools | 71.8β Thinking High Β· no tools | +10.3 pts#4 vs #53 of 106 |
| LiveBench 2.0Cross-capability Suites Β· Overall score (equal-weight mean of 7 suite scores, 0-100) | 86.1Max Β· task tools | 47.7β Thinking High Β· tools | 1.81Γ#3 vs #47 of 51 |
| Vals Index v2.1Capability Indices Β· GDP-weighted accuracy (%) | 63.1Max Β· task tools | 48.0β Thinking High Β· tools | +15.1 pts#6 vs #30 of 38 |
Repository Engineering Β· Composite index (0β100)
Maintenance & Optimization Β· Hidden-test pass rate (%)
Repository Engineering Β· Pass@1 (DeepSWE v1.1)
Code Generation & Editing Β· Weighted rubric score (%)
Not counted Β· Different or unrecorded test environments β excluded from the tally
Repository Engineering Β· Weighted rubric score
Algorithmic Coding Β· Subtask points (%)
Repository Engineering Β· Task Resolve Rate (%)
Agentic Development Β· Resolution rate (%)
Not counted Β· Different or unrecorded test environments β excluded from the tally
Agentic Development Β· Resolution rate (%)
Agentic Development Β· Accuracy (%)
Code Generation & Editing Β· Pass rate (%)
ML Engineering Β· Official score (0-1; higher is better)
Mathematics Β· Accuracy (verification_code)
Mathematics Β· Accuracy (verification_code)
Mathematics Β· Accuracy (%)
Scientific Reasoning Β· Accuracy (%)
Biology & Genomics Β· Accuracy (%)
Clinical Workflows Β· Accuracy (%)
Clinical Workflows Β· Rubric pass rate (%)
Visual Understanding Β· Accuracy
Finance Β· Accuracy (%)
Finance Β· Score
Legal Β· Accuracy (%)
Legal Β· Score
Finance Β· Accuracy (%, rubric Γ citation quality)
Capability Indices Β· Index (0-100)
Cross-capability Suites Β· Accuracy
Cross-capability Suites Β· Overall score (equal-weight mean of 7 suite scores, 0-100)
Capability Indices Β· GDP-weighted accuracy (%)
A dash means no result is available.
Time-of-day providers use peak-hour rates; fixed-price providers use standard rates. Off-peak discounts remain in the published details.
Drag the handle above the bar for cache hits, the one below it for the input : output mix Β· 5% steps
Blended price = effective input rate at 90% cache hits Γ 80% + output rate Γ 20%
GPT-6 Astra costs2,088%morethan Mistral Large 4 per million tokens β 21.9Γ the price.Using listed rates vs standard rates at this token mix.
GPT-6 Astra: Standard pricing. context_length <= 272000 OpenAI official pricing (developers.openai.com/api/docs/pricing) verified 2026-11-13; standard tier; <=272K input tokens context_length > 272000 OpenAI official pricing (developers.openai.com/api/docs/pricing) verified 2026-11-13; standard tier; >272K input tokens
Mistral Large 4: Standard pricing. Mistral official pricing (docs.mistral.ai/inference/pricing) verified 2026-11-13; price reduced from $1.36 Mistral official pricing (docs.mistral.ai/inference/pricing) verified 2026-11-13; price reduced from $4.18
| Price / 1M tokens | GPT-6 Astra | Mistral Large 4 | Ratio |
|---|---|---|---|
| Input | $10.00 | $0.70 | 14.3Γ |
| Output | $50.00 | $2.10 | 23.8Γ |
| Cached input | $1.00 | $0.07 | 14.3Γ |
| Cache write | $12.50 | β | β |
| Blended | $11.52 | $0.526 | 21.9Γ |
Estimates use the selected token mix. Cache writes, tool fees and long-context surcharges are excluded.
Reported cost per task, paired with the score from the same run. Settings and sources may differ; costs do not affect the verdict.
GPT-6 Astra relative to Mistral Large 4
Summary: 4 of 8 shared benchmarks have matching recorded settings and sources. The average is a geometric mean of 4 positive-cost ratios, one equal weight per benchmark. This benchmark set can include related tests and indexes; it is not a workload cost estimate.
GPT-6 Astra 4.5Γ more expensive
GPT-6 Astra 3.5Γ more expensive
GPT-6 Astra 2.8Γ more expensive
GPT-6 Astra 1.9Γ more expensive
Settings or sources differ
Settings or sources differ
Settings or sources differ
Settings or sources differ
GPT-6 Astra 4.5Γ more expensive
GPT-6 Astra 3.5Γ more expensive
GPT-6 Astra 2.8Γ more expensive
GPT-6 Astra 1.9Γ more expensive
Settings or sources differ
Settings or sources differ
Settings or sources differ
Settings or sources differ
Effort, protocol, aggregation and source match; other run conditions may differ.
GPT-6 Astra
Mistral Large 4
Effort, protocol, aggregation and source match; other run conditions may differ.
GPT-6 Astra
Mistral Large 4
Settings or sources differ or are not recorded; interpret the comparison with care.
GPT-6 Astra
Mistral Large 4
Effort, protocol, aggregation and source match; other run conditions may differ.
GPT-6 Astra
Mistral Large 4
Effort, protocol, aggregation and source match; other run conditions may differ.
GPT-6 Astra
Mistral Large 4
Settings or sources differ or are not recorded; interpret the comparison with care.
GPT-6 Astra
Mistral Large 4
Settings or sources differ or are not recorded; interpret the comparison with care.
GPT-6 Astra
Mistral Large 4
Settings or sources differ or are not recorded; interpret the comparison with care.
GPT-6 Astra
Mistral Large 4
| Benchmark | GPT-6 Astra | Mistral Large 4 |
|---|---|---|
| AA Coding Agent Index v2.0 | Not reported | |
| AA Intelligence Index v4.3 | 2 more settings | Not reported |
| AA-Omniscience | 2 more settings | Not reported |
| Agent Arena | $5.4265 / task Median Β· Net improvement index (percentage points vs. average model): 14.6 Max (With Tools + Internet) | Not reported |
| Agent Arena Index | $4.7264 / task Median Β· Agent Arena Index rating (index points): 1282 Max (With Tools + Internet) | Not reported |
| Agents' Last Exam v1.0 | 3 more settings | Not reported |
| ARC-AGI-1 | 5 more settings$0.2837 / task Average Β· Score (%): 98.5 Thinking High (With Tools) $0.2265 / task Average Β· Score (%): 96.5 Thinking Low (With Tools) $0.2398 / task Average Β· Score (%): 97.5 Thinking Medium (With Tools) | Not reported |
| ARC-AGI-2 | $0.6681 / task Average Β· ARC-AGI-2 Score: 92.1 Thinking High (With Tools) 3 more settings$0.4162 / task Average Β· ARC-AGI-2 Score: 85.4 Thinking Low (With Tools) $0.4801 / task Average Β· ARC-AGI-2 Score: 92.1 Thinking Medium (With Tools) $0.8287 / task Average Β· ARC-AGI-2 Score: 93.3 Extra-High (With Tools) | Not reported |
| AutomationBench | $1.44 / task Average Β· Pass Rate: 37.1 Thinking High (With Tools) Β· zapier-private-heldout 3 more settings$1.08 / task Average Β· Pass Rate: 30.3 Thinking Low (With Tools) Β· zapier-private-heldout $1.27 / task Average Β· Pass Rate: 34.1 Thinking Medium (With Tools) Β· zapier-private-heldout $1.50 / task Average Β· Pass Rate: 39.0 Extra-High (With Tools) Β· zapier-private-heldout | Not reported |
| BenchCAD 2.1 Core | $9.71 / task Average Β· Core mean geometry/circuit score (0β100): 62.6 Max (With Tools) $5.38 / task Average Β· Core mean geometry/circuit score (0β100): 51.1 Thinking High (With Tools) $1.14 / task Average Β· Core mean geometry/circuit score (0β100): 20.3 Thinking Low (With Tools) 2 more settings$3.73 / task Average Β· Core mean geometry/circuit score (0β100): 38.7 Thinking Medium (With Tools) $6.32 / task Average Β· Core mean geometry/circuit score (0β100): 53.9 Extra-High (With Tools) | Not reported |
| BioMysteryBench | Not reported | |
| Code Migration | Not reported | |
| CWE-bench v1 | Not reported | |
| DeepSWE | 2 more settings | Not reported |
| EMB (Excel Modeling) | Not reported | |
| Finance Agent v2 | Not reported | |
| FrontierCode 1.1 | $4.586 / task Average Β· Weighted rubric score: 53.3 Max (With Tools) Β· frontiercode-1-1-main $3.0149 / task Average Β· Weighted rubric score: 50.9 Thinking High (With Tools) Β· frontiercode-1-1-main $1.70 / task Average Β· Weighted rubric score: 45.3 Thinking Low (With Tools) Β· frontiercode-1-1-main 2 more settings$2.4259 / task Average Β· Weighted rubric score: 48.8 Thinking Medium (With Tools) Β· frontiercode-1-1-main $3.2755 / task Average Β· Weighted rubric score: 50.6 Extra-High (With Tools) Β· frontiercode-1-1-main | Not reported |
| FrontierCS 2 | Not reported | |
| FrontierSWE v2 | Not reported | |
| GDPx | Not reported | |
| Harvey's Legal Agent Benchmark | Not reported | |
| HLE | $0.0389 / task Average Β· Accuracy: 49.2 Thinking Low (No Tools) Β· Text only Β· text-only $0.1005 / task Average Β· Accuracy: 52.7 Thinking Medium (No Tools) Β· Text only Β· text-only 2 more settings$0.1618 / task Average Β· Accuracy: 53.1 Thinking High (No Tools) Β· Text only Β· text-only $0.2532 / task Average Β· Accuracy: 54.6 Extra-High (No Tools) Β· Text only Β· text-only | Not reported |
| IOI (Vals v2) | Not reported | |
| Legal Research Bench | Not reported | |
| MedCode | Not reported | |
| MedScribe | Not reported | |
| MysteryMechanism | Not reported | |
| ProofBench v1.1 (Lean 4) | Not reported | |
| SWE-sweep | Not reported | |
| Tax Agent Bench | Not reported | $2.7212 / task Average Β· Accuracy (%, rubric Γ citation quality): 63.3 Thinking High (With Tools) |
| Terminal-Bench-Science 0.1 | $20.80 / task Average Β· Task success rate (%): 62.9 Unspecified Reasoning (With Tools) | Not reported |
| Vals Index v2.2 | Not reported | |
| Vibe Code Bench v1.1 | Not reported |
| Attribute | GPT-6 Astra | Mistral Large 4 |
|---|---|---|
| Availability | Preview | Preview |
| Provider | OpenAI | MistralAI |
| Model type | Reasoning model | Multimodal model |
| First release | September 3, 2026 | October 6, 2026 |
| Context window | 1.05M tokens | 1M tokens |
| Max output | 128K tokens | β |
| Total parameters | β | 1.1T |
| Active parameters | β | 49B |
| Architecture | Undisclosed | Mixture of Experts (MoE) |
| Documented capabilities | Reasoning, Multilingual, Coding, Function calling, Tool use, Structured output, Computer use, Web search, Long context | Reasoning, Multilingual, Function calling, Tool use, Structured output, Long context, Coding |
| Knowledge cutoff | 2026-04-30 | β |
| Input modalities | Text, Image | Text, Image |
| Output modalities | Text | Text |
| Weights | Closed / API only | Closed / API only |
| Commercial use | Proprietary | Proprietary |
These are reported-score comparisons, not controlled head-to-head tests.
What people found comparing these models, or using either one. Comments also appear on the models' own pages.
Used GPT-6 Astra or Mistral Large 4? Sign in with Google to share what worked, what did not, and which API or setup you tested.
No community comments on either model yet.
Alternatives with the most benchmark overlap for this matchup.