Hy3 vs Hy3 preview
Different efforts: thinking · non-thinking
Humanity's Last Exam Humanity's Last Exam: Very hard expert questions across many fields, written to stump today’s models
Hy333.5
Hy3 preview7
Terminal-Bench 4.0 Terminal-Bench 4.0 · source setting 2: Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite
Hy30.5
Hy3 preview—
AutomationBench-AA AutomationBench-AA: Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation
Hy318.4
Hy3 preview—
Arena Text · creative writing Arena Text · creative writing: Human pairwise preferences in the stated text vote pool
Hy31,422
Hy3 preview1,354
| Benchmark | Hy3 | Hy3 preview |
|---|---|---|
| Humanity's Last Exam | 33.5 | 7 |
| Terminal-Bench 4.0 | 0.5 | — |
| AutomationBench-AA | 18.4 | — |
| Arena Text · creative writing | 1,422 | 1,354 |
Specs
| Spec | Hy3 | Hy3 preview |
|---|---|---|
| Input, per 1M | $0.13 | $0.18 |
| Output, per 1M | $0.53 | $0.60 |
| Cost per task? | $0.072 | — |
| Speed, tok/s | 85 | — |
| Context | 256K | 256K |
| Weights | Open | Open |
| Released | Jul 2026 | Apr 2026 |
| Memory (4-bit) | 179 GB | 179 GB |
| Licence | Apache-2.0 | custom licence |
| Commercial use | Allowed | Restricted |
| See images | — | — |
| Read PDFs | — | — |
| Hear audio | — | — |
| Watch video | — | — |
| Call tools | ✓ | ✓ |
| Force a tool call | Partial | Partial |
| Return strict JSON | ✓ | — |
| Search the web | ? | ? |
| Run code | ? | ? |
| Use a computer | ? | ? |
What breaks when you switch
Hy3 → Hy3 preview
No strict JSON
Hy3 preview → Hy3
Nothing known.