GPT-6 Models Compared: Every Variant Benchmarked (Nov 2026)
As of Nov 3, 2026, isitsota tracks 7 GPT-6 models: GPT-6.1 Terra, GPT-6.1 Sol, GPT-6 Luna, GPT-6 Sol, GPT-6 Astra, GPT-6 Astra Law, and GPT-6 Rosalind. GPT-6.1 Sol is state of the art in Vision (shared) since Sep 30, 2026. GPT-6 Astra is state of the art in Math since Sep 28, 2026 and Vision (shared) since Sep 30, 2026. GPT-6 Astra Law and GPT-6 Rosalind have no independent benchmark results yet.
Last updated Nov 3, 2026
Models in the family
| Model | Released | Availability | Context | Price | Crowns | Best rank |
|---|---|---|---|---|---|---|
| GPT-6.1 Terra | Oct 26, 2026 | ga | 1,050,000 | $2 / 1M tokens (blended) | — | #8 Science |
| GPT-6.1 Sol | Sep 29, 2026 | ga | 1,050,000 | $4 / 1M tokens (blended) | Vision | #1 Vision |
| GPT-6 Luna | Sep 22, 2026 | ga | 1,050,000 | $0.2 / 1M tokens (blended) | — | #17 Coding › Agentic |
| GPT-6 Sol | Sep 22, 2026 | ga | 1,050,000 | $4 / 1M tokens (blended) | — | #5 Coding › Agentic |
| GPT-6 Astra | Sep 3, 2026 | ga | 1,050,000 | $20 / 1M tokens (blended) | Math, Vision | #1 Math |
| GPT-6 Astra Law | — | announced | — | — | Pending benchmarks | — |
| GPT-6 Rosalind | — | announced | — | — | Pending benchmarks | — |
Rank by category
| Category | GPT-6.1 Terra | GPT-6.1 Sol | GPT-6 Luna | GPT-6 Sol | GPT-6 Astra |
|---|---|---|---|---|---|
| General LLM | #67 | #25 | #105 | #54 | #27 |
| Reasoning | — | #17 | #74 | #49 | #29 |
| Math | — | — | — | — | #1 |
| Science | #8 | — | #22 | — | #1 |
| Coding | — | #3 | #29 | #9 | #2 |
| Coding › Agentic | — | #2 | #17 | #5 | #2 |
| Coding › Web dev | #13 | #5 | #36 | #11 | #2 |
| Long context | — | #20 | #75 | #52 | #24 |
| Vision | — | #1 | #17 | #9 | #1 |
| Factuality | — | #34 | #28 | #83 | #32 |
| Cybersecurity | — | #2 | #18 | #5 | #1 |
| Legal | #35 | #20 | #30 | #32 | #18 |
| Healthcare | — | — | #41 | #36 | — |
Overall rank on each category board. A dash means the model is not ranked there.
Benchmarks side by side
| Benchmark | GPT-6.1 Terra | GPT-6.1 Sol | GPT-6 Luna | GPT-6 Sol | GPT-6 Astra |
|---|---|---|---|---|---|
| ARC-AGI | — | 98.5% | 86.7% | 95.5% | 98.5% |
| Code Migration | — | 77.3 | 46.7 | — | 80.6 |
| CyberBench | — | 39.3 | 76.3 | 78.0 | 41.1 |
| Excel Modeling Benchmark (EMB) | 73.3 | 71.1 | 70.0 | — | 72.7 |
| Finance Agent | 56.0 | 61.8 | 56.8 | — | 60.4 |
| GDP.pdf | — | 35.0% | 23.0% | 26.4% | 34.2% |
| GPQA Diamond | 94.3% | 95.4% | 90.5% | 94.3% | 95.8% |
| Legal Research Bench | 26.0 | 38.5 | 30.3 | 28.8 | 39.4 |
| LMArena Text (style-controlled, coding) | — | 1541 | 1506 | 1519 | 1542 |
| LMArena Text (style-controlled, hard prompts) | — | 1506 | 1464 | 1484 | 1500 |
| LMArena Text (style-controlled, longer query) | — | 1491 | 1451 | 1467 | 1488 |
| LMArena Text (style-controlled, math) | — | 1509 | 1447 | 1448 | 1489 |
| LMArena Text (style-controlled, overall) | — | 1481 | 1441 | 1458 | 1476 |
| LMArena Text factuality | — | 1466 | 1469 | 1440 | 1467 |
| LMArena Vision (style-controlled) | — | 1294 | 1240 | 1266 | 1285 |
Head-to-head pages
Which GPT-6 model is best?
GPT-6 Astra: as of Nov 3, 2026, it holds 2 category crowns (Math since Sep 28, 2026 and Vision (shared) since Sep 30, 2026).
Is GPT-6 state of the art?
Yes, in 2 categories as of Nov 3, 2026: GPT-6.1 Sol in Vision (shared) since Sep 30, 2026; GPT-6 Astra in Math since Sep 28, 2026 and Vision (shared) since Sep 30, 2026.
When was GPT-6 released?
The first GPT-6 model on isitsota, GPT-6 Astra, was released on Sep 3, 2026. The newest, GPT-6.1 Terra, was released on Oct 26, 2026.
Which GPT-6 model is cheapest?
GPT-6 Luna, at $0.2 / 1M tokens (blended). The most expensive tracked variant is GPT-6 Astra ($20 / 1M tokens (blended)).
How often is this page updated?
Rankings are recomputed automatically when new benchmark results arrive; this page reflects data as of Nov 3, 2026.