Model comparison
GPT-6 Astra vs Phi-4
GPT-6 Astra is the stronger model overall, scoring 70.8 to 31.2 on the Noometry Index. Phi-4 costs 229× less per token, which makes it the better buy when GPT-6 Astra's lead doesn't matter for your workload.
Last verified . 23 shared benchmarks.
MicrosoftSummary
- They share 23 benchmarks with published results for both. GPT-6 Astra scores higher in 9 categories and Phi-4 in 0 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in math, where GPT-6 Astra leads 93.5 to 20.8.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 100% for GPT-6 Astra and 13.8% for Phi-4.
- Phi-4 is cheaper at $0.07 / $0.14 per million input/output tokens, against $10 / $50 for GPT-6 Astra.
- GPT-6 Astra accepts more context: 1.05M tokens versus 128K.
- Phi-4 has downloadable open weights; the other is API-only.
Side by side
| GPT-6 Astra | Phi-4 | |
|---|---|---|
| Provider | OpenAI | Microsoft |
| Noometry Index | 70.8 | 31.2 |
| Released | 2026-09-03 | 2024-12-11 |
| Weights | Proprietary | Open |
| Context window | 1.05M | 128K |
| Max output | 128K | 4K |
| Input $ / M tokens | $10 | $0.07 |
| Output $ / M tokens | $50 | $0.14 |
| Results tracked | 56 | 37 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-6 Astra leads
GPT-6 Astra: 73.7 (#2), Phi-4: 34.4 (#239)
| Benchmark | GPT-6 Astra | Phi-4 |
|---|---|---|
| LMArena Coding | 1487 | 1231 |
| DeepSWE | 74.1% | — |
| FrontierCode | 53.3% | — |
| LMArena WebDev | 1786 | — |
| FrontierSWE | 65.5% | — |
| SciCode | 56.5% | — |
| GSO | 79.4% | — |
| WeirdML | 93.6% | — |
| BigCodeBench Instruct | — | 45.5% |
| LiveBench Coding | — | 30.7% |
| MirrorCode | 46.7% | — |
| BigCodeBench Complete | — | 55.4% |
| ALE-Bench | 2,951 | — |
Agentic & Tool Use GPT-6 Astra leads
GPT-6 Astra: 52.9 (#3), Phi-4: 22.8 (#128)
| Benchmark | GPT-6 Astra | Phi-4 |
|---|---|---|
| BALROG | 68.3% | 11.6% |
| APEX-Agents | 64.7% | — |
| Berkeley Function Calling Leaderboard | — | 28.8% |
| Remote Labor Index | 20.8% | — |
| GDP.pdf | 34.2% | — |
| Vending-Bench 2 | 15,515 | — |
Reasoning GPT-6 Astra leads
GPT-6 Astra: 85.1 (#1), Phi-4: 17.7 (#291)
| Benchmark | GPT-6 Astra | Phi-4 |
|---|---|---|
| Chess Puzzles | 72% | 1% |
| LMArena Hard Prompts | 1462 | 1220 |
| Epoch Capabilities Index | 166.45 | 130.42 |
| ARC-AGI-2 | 95% | — |
| NYT Connections (extended) | 98.1% | — |
| ARC-AGI-1 | 98.5% | — |
| CritPt | 31.7% | — |
| EBR-Bench | 76.2% | — |
| LiveBench Reasoning | — | 47.8% |
| Mystery Game Puzzles | 84% | — |
| DTBench | 97.3% | — |
| LiveBench Data Analysis | — | 45.2% |
| LMCA | 64.4% | — |
| Bench to the Future 3 | 0.14 | — |
| LiveBench | — | 41.6% |
Math GPT-6 Astra leads
GPT-6 Astra: 93.5 (#2), Phi-4: 20.8 (#285)
| Benchmark | GPT-6 Astra | Phi-4 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 100% | 13.8% |
| LMArena Math | 1465 | 1246 |
| FrontierMath (Tiers 1-3) | 93.7% | — |
| FrontierMath Tier 4 | 97.6% | — |
| ProofBench | 99% | — |
| LiveBench Math | — | 42% |
| MATH Level 5 | — | 64.9% |
| FrontierMath Erdős | 2.9% | — |
Knowledge GPT-6 Astra leads
GPT-6 Astra: 75.3 (#1), Phi-4: 32.6 (#209)
| Benchmark | GPT-6 Astra | Phi-4 |
|---|---|---|
| GPQA Diamond | 95.8% | 56.1% |
| Vectara Hallucination Rate | 8.7% | 3.7% |
| LMArena Expert | 1483 | 1203 |
| Humanity's Last Exam | 54.8% | — |
| SimpleQA Verified | 75.6% | — |
| Confabulations | — | 29.4% |
| MMLU | — | 84.8% |
Multimodal Not comparable
GPT-6 Astra: 55.0 (#3), Phi-4: —
| Benchmark | GPT-6 Astra | Phi-4 |
|---|---|---|
| LMArena Vision | 1281 | — |
| Blueprint-Bench 2 | 49.7% | — |
| Furniture Assembly | 80% | — |
| LMArena Document | 1468 | — |
Multilingual GPT-6 Astra leads
GPT-6 Astra: 53.7 (#61), Phi-4: 37.2 (#237)
| Benchmark | GPT-6 Astra | Phi-4 |
|---|---|---|
| LMArena Non-English | 1430 | 1197 |
| LMArena Chinese | 1484 | 1212 |
| LMArena French | 1456 | 1224 |
| LMArena German | 1440 | 1222 |
| LMArena Japanese | 1379 | 1158 |
| LMArena Korean | 1426 | 1151 |
| LMArena Russian | 1436 | 1209 |
| LMArena Spanish | 1407 | 1234 |
Instruction Following GPT-6 Astra leads
GPT-6 Astra: 76.3 (#44), Phi-4: 60.4 (#251)
| Benchmark | GPT-6 Astra | Phi-4 |
|---|---|---|
| LMArena Instruction Following | 1450 | 1201 |
| LiveBench Instruction Following | — | 58.4% |
Long Context GPT-6 Astra leads
GPT-6 Astra: 44.5 (#62), Phi-4: 36.9 (#226)
| Benchmark | GPT-6 Astra | Phi-4 |
|---|---|---|
| LMArena Longer Query | 1456 | 1217 |
Writing & Preference GPT-6 Astra leads
GPT-6 Astra: 75.3 (#7), Phi-4: 40.5 (#244)
| Benchmark | GPT-6 Astra | Phi-4 |
|---|---|---|
| LMArena Text | 1441 | 1217 |
| LMArena Creative Writing | 1418 | 1182 |
| LMArena Multi-Turn | 1448 | 1206 |
| Short-Story Creative Writing | — | 62.6% |
| EQ-Bench Creative Writing | 2173 | — |
| LiveBench Language | — | 25.6% |
Frequently asked questions
Is GPT-6 Astra better than Phi-4?
GPT-6 Astra is the stronger model overall, scoring 70.8 to 31.2 on the Noometry Index. Phi-4 costs 229× less per token, which makes it the better buy when GPT-6 Astra's lead doesn't matter for your workload.
Which is cheaper, GPT-6 Astra or Phi-4?
Phi-4 is cheaper. It lists at $0.07 per million input tokens and $0.14 per million output tokens; GPT-6 Astra lists at $10 and $50.
Is GPT-6 Astra or Phi-4 better for coding?
GPT-6 Astra scores higher on coding benchmarks: 73.7 versus 34.4 in the Noometry coding category.
Which has the bigger context window?
GPT-6 Astra does, with 1.05M tokens against 128K.
How many benchmarks do GPT-6 Astra and Phi-4 share?
23 benchmarks have published results for both models. GPT-6 Astra has 56 scored results on Noometry and Phi-4 has 37.