Model comparison
DeepSeek V4 Flash vs GPT-6 Astra
GPT-6 Astra is the stronger model overall, scoring 70.8 to 53.6 on the Noometry Index. DeepSeek V4 Flash costs 76× less per token, which makes it the better buy when GPT-6 Astra's lead doesn't matter for your workload.
Last verified . 38 shared benchmarks.
DeepSeekSummary
- They share 38 benchmarks with published results for both. DeepSeek V4 Flash scores higher in 0 categories and GPT-6 Astra in 8 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in math, where GPT-6 Astra leads 93.5 to 60.3.
- The biggest single-benchmark swing is FrontierMath Tier 4: 24.4% for DeepSeek V4 Flash and 97.6% for GPT-6 Astra.
- DeepSeek V4 Flash is cheaper at $0.15 / $0.60 per million input/output tokens, against $10 / $50 for GPT-6 Astra.
- GPT-6 Astra accepts more context: 1.05M tokens versus 1M.
- DeepSeek V4 Flash has downloadable open weights; the other is API-only.
Side by side
| DeepSeek V4 Flash | GPT-6 Astra | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 53.6 | 70.8 |
| Released | 2026-04-24 | 2026-09-03 |
| Weights | Open | Proprietary |
| Context window | 1M | 1.05M |
| Max output | 393K | 128K |
| Input $ / M tokens | $0.15 | $10 |
| Output $ / M tokens | $0.60 | $50 |
| Results tracked | 41 | 56 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-6 Astra leads
DeepSeek V4 Flash: 47.9 (#59), GPT-6 Astra: 73.7 (#2)
| Benchmark | DeepSeek V4 Flash | GPT-6 Astra |
|---|---|---|
| FrontierCode | 18.8% | 53.3% |
| LMArena WebDev | 1582 | 1786 |
| SciCode | 49.9% | 56.5% |
| WeirdML | 63% | 93.6% |
| LMArena Coding | 1457 | 1487 |
| ALE-Bench | 1,306 | 2,951 |
| DeepSWE | — | 74.1% |
| FrontierSWE | — | 65.5% |
| GSO | — | 79.4% |
| MirrorCode | — | 46.7% |
Agentic & Tool Use Not comparable
DeepSeek V4 Flash: —, GPT-6 Astra: 52.9 (#3)
| Benchmark | DeepSeek V4 Flash | GPT-6 Astra |
|---|---|---|
| APEX-Agents | — | 64.7% |
| Remote Labor Index | — | 20.8% |
| BALROG | — | 68.3% |
| GDP.pdf | — | 34.2% |
| Vending-Bench 2 | — | 15,515 |
Reasoning GPT-6 Astra leads
DeepSeek V4 Flash: 53.7 (#30), GPT-6 Astra: 85.1 (#1)
| Benchmark | DeepSeek V4 Flash | GPT-6 Astra |
|---|---|---|
| ARC-AGI-2 | 61.4% | 95% |
| NYT Connections (extended) | 89.6% | 98.1% |
| ARC-AGI-1 | 89% | 98.5% |
| CritPt | 16.6% | 31.7% |
| Chess Puzzles | 33% | 72% |
| LMArena Hard Prompts | 1444 | 1462 |
| Mystery Game Puzzles | 34% | 84% |
| DTBench | 90.9% | 97.3% |
| LMCA | 41.7% | 64.4% |
| Epoch Capabilities Index | 154.49 | 166.45 |
| SimpleBench | 61.1% | — |
| Kagi LLM Benchmark | 52.2% | — |
| EBR-Bench | — | 76.2% |
| Bench to the Future 3 | — | 0.14 |
Math GPT-6 Astra leads
DeepSeek V4 Flash: 60.3 (#37), GPT-6 Astra: 93.5 (#2)
| Benchmark | DeepSeek V4 Flash | GPT-6 Astra |
|---|---|---|
| FrontierMath (Tiers 1-3) | 57.5% | 93.7% |
| FrontierMath Tier 4 | 24.4% | 97.6% |
| OTIS Mock AIME 2024-2025 | 94.4% | 100% |
| ProofBench | 56% | 99% |
| LMArena Math | 1427 | 1465 |
| MathArena Final-Answer Competitions | 76.5% | — |
| FrontierMath Erdős | — | 2.9% |
Knowledge GPT-6 Astra leads
DeepSeek V4 Flash: 55.4 (#48), GPT-6 Astra: 75.3 (#1)
| Benchmark | DeepSeek V4 Flash | GPT-6 Astra |
|---|---|---|
| GPQA Diamond | 91% | 95.8% |
| SimpleQA Verified | 33.6% | 75.6% |
| LMArena Expert | 1441 | 1483 |
| Humanity's Last Exam | — | 54.8% |
| Vectara Hallucination Rate | — | 8.7% |
Multimodal Not comparable
DeepSeek V4 Flash: —, GPT-6 Astra: 55.0 (#3)
| Benchmark | DeepSeek V4 Flash | GPT-6 Astra |
|---|---|---|
| LMArena Vision | — | 1281 |
| Blueprint-Bench 2 | — | 49.7% |
| Furniture Assembly | — | 80% |
| LMArena Document | — | 1468 |
Multilingual Too close to call
DeepSeek V4 Flash: 53.0 (#72), GPT-6 Astra: 53.7 (#61)
| Benchmark | DeepSeek V4 Flash | GPT-6 Astra |
|---|---|---|
| LMArena Non-English | 1420 | 1430 |
| LMArena Chinese | 1468 | 1484 |
| LMArena French | 1439 | 1456 |
| LMArena German | 1418 | 1440 |
| LMArena Japanese | 1406 | 1379 |
| LMArena Korean | 1384 | 1426 |
| LMArena Russian | 1428 | 1436 |
| LMArena Spanish | 1436 | 1407 |
Instruction Following GPT-6 Astra leads
DeepSeek V4 Flash: 74.9 (#81), GPT-6 Astra: 76.3 (#44)
| Benchmark | DeepSeek V4 Flash | GPT-6 Astra |
|---|---|---|
| LMArena Instruction Following | 1421 | 1450 |
Long Context Too close to call
DeepSeek V4 Flash: 43.8 (#85), GPT-6 Astra: 44.5 (#62)
| Benchmark | DeepSeek V4 Flash | GPT-6 Astra |
|---|---|---|
| LMArena Longer Query | 1434 | 1456 |
Writing & Preference GPT-6 Astra leads
DeepSeek V4 Flash: 63.8 (#61), GPT-6 Astra: 75.3 (#7)
| Benchmark | DeepSeek V4 Flash | GPT-6 Astra |
|---|---|---|
| LMArena Text | 1432 | 1441 |
| LMArena Creative Writing | 1403 | 1418 |
| EQ-Bench Creative Writing | 1559 | 2173 |
| LMArena Multi-Turn | 1449 | 1448 |
Frequently asked questions
Is DeepSeek V4 Flash better than GPT-6 Astra?
GPT-6 Astra is the stronger model overall, scoring 70.8 to 53.6 on the Noometry Index. DeepSeek V4 Flash costs 76× less per token, which makes it the better buy when GPT-6 Astra's lead doesn't matter for your workload.
Which is cheaper, DeepSeek V4 Flash or GPT-6 Astra?
DeepSeek V4 Flash is cheaper. It lists at $0.15 per million input tokens and $0.60 per million output tokens; GPT-6 Astra lists at $10 and $50.
Is DeepSeek V4 Flash or GPT-6 Astra better for coding?
GPT-6 Astra scores higher on coding benchmarks: 73.7 versus 47.9 in the Noometry coding category.
Which has the bigger context window?
GPT-6 Astra does, with 1.05M tokens against 1M.
How many benchmarks do DeepSeek V4 Flash and GPT-6 Astra share?
38 benchmarks have published results for both models. DeepSeek V4 Flash has 41 scored results on Noometry and GPT-6 Astra has 56.