GPT-5.6 Sol vs GPT-5.6 Terra: Specs & Benchmark Comparison
GPT-5.6 Sol and GPT-5.6 Terra both come from OpenAI, and both were released in July 2026.
The two models share 45 published benchmarks. GPT-5.6 Sol leads on 40 of them, GPT-5.6 Terra on 3. The widest gaps are on ExploitBench, where GPT-5.6 Sol scores 73.5% against 52.9%; Terminal-Bench 4.0, where GPT-5.6 Sol scores 37.3% against 21.5%. Averaged across everything we track, GPT-5.6 Sol sits at 63.9% and GPT-5.6 Terra at 59.0%.
GPT-5.6 Terra is the cheaper API at $2 per million input tokens and $12 per million output tokens, roughly 3 times cheaper than GPT-5.6 Sol at $5 and $30. Both accept a context window of about 1M tokens. Both accept text and images as input.
| Characteristic | GPT-5.6 Sol | GPT-5.6 Terra |
|---|---|---|
| Company | OpenAI | OpenAI |
| Release Date | July 9, 2026 | July 9, 2026 |
| Parameters | — | — |
| Multimodal | Yes | Yes |
| Context (input) | 1.1M | 1.1M |
| Context (output) | 128K | 128K |
| Input Price / 1M | $5.00 | $2.00 |
| Output Price / 1M | $30.00 | $12.00 |
| Average Score | 63.9% | 59.0% |
| Benchmarks | ||
| ExploitBench | 73.5% | 52.9% |
| Terminal-Bench 4.0 | 37.3% | 21.5% |
| FrontierMath Tier 4 (v2) | 83.0% | 68.3% |
| Graphwalks BFS >128k | 90.7% | 76.9% |
| SEC-bench Pro | 71.2% | 57.7% |
| MedChemBench (Internal) | 48.3% | 35.0% |
| OSWorld 2.0 | 62.6% | 50.2% |
| KernelGen 1P | 61.1% | 49.2% |
| ExploitGym | 33.7% | 23.2% |
| BenchCAD | 70.6% | 62.3% |
| ARC-AGI-3 | 7.8% | 0.8% |
| FrontierCode 1.1 | 47.5% | 41.3% |
| GDP.pdf | 30.7% | 24.7% |
| Management Consulting Tasks (Internal) | 43.2% | 37.2% |
| Graphwalks BFS 1M | 77.1% | 71.2% |
| GeneBench-Pro | 28.7% | 23.3% |
| BenchCAD (with Python tool) | 83.4% | 78.2% |
| Capture-the-Flag Challenges (Internal) | 96.7% | 91.8% |
| Toolathlon | 58.0% | 53.1% |
| NanoGPT | 9.7% | 14.5% |
| FrontierMath | 89.0% | 84.9% |
| Artificial Analysis | 59.0% | 55.0% |
| LifeSciBench | 59.9% | 56.0% |
| Search and Function-Calling | 91.0% | 94.6% |
| DeepSWE | 72.7% | 69.6% |
| DeepSWE 1.1 | 73.0% | 70.0% |
| BrowseComp | 90.4% | 87.5% |
| AutomationBench | 18.1% | 15.2% |
| HealthBench Professional | 60.5% | 57.7% |
| MMMU-Pro (with tools) | 84.6% | 82.0% |
| Agents' Last Exam | 52.7% | 50.4% |
| MMMU-Pro | 83.0% | 80.7% |
| Big Finance Bench | 53.0% | 51.0% |
| MRCR v2 (8-needle) | 91.5% | 89.6% |
| GPQA | 94.6% | 92.9% |
| RSI Index | 57.9% | 56.3% |
| Terminal-Bench 2.1 | 88.8% | 87.4% |
| MRCR v2 (8-needle, 512K-1M) | 73.8% | 72.5% |
| PostTrainBench Lite | 50.3% | 51.5% |
| SWE-Bench Pro | 64.6% | 63.4% |
| Internal Research Debugging Evaluation | 68.3% | 67.8% |
| HealthBench Consensus | 95.5% | 95.1% |
| HealthBench Hard | 33.1% | 32.7% |
| Connectors | 100.0% | 100.0% |
| HealthBench | 57.0% | 57.0% |
Visual Benchmark Comparison
Verdict
GPT-5.6 Terra leads in 1 out of 4 comparison categories.
Both models show comparable average scores: GPT-5.6 Sol — 0.6, GPT-5.6 Terra — 0.6.
GPT-5.6 Terra is 2.5x cheaper: input $2.00/1M vs $5.00/1M tokens.
Same context size: 1M tokens.
Both models were released around the same time: 7/9/2026 and 7/9/2026.
More About These Models
Frequently Asked Questions
Which is better for coding — GPT-5.6 Sol or GPT-5.6 Terra?
Which model is cheaper — GPT-5.6 Sol or GPT-5.6 Terra?
Which has a larger context window — GPT-5.6 Sol or GPT-5.6 Terra?
The GPT-5.6 Sol and GPT-5.6 Terra comparison is updated for 2026. Data includes benchmark results, API pricing, context window size and other specifications. For more detailed information, visit the GPT-5.6 Sol or GPT-5.6 Terra page. See also the complete list of AI model comparisons.