GPT-6 Astra vs MiniMax M3.1 Flash
Different efforts: medium · thinking
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
GPT-6 Astra49.5
MiniMax M3.1 Flash25.8
FrontierCode 1.1 FrontierCode 1.1: Agents implement changes in real repositories, graded against maintainer-defined requirements
GPT-6 Astra · max47
MiniMax M3.1 Flash38.6
WebDev Arena WebDev Arena: Human pairwise preferences in the stated webdev vote pool
GPT-6 Astra · max1,774
MiniMax M3.1 Flash1,587
SciCode SciCode: Writing code for real scientific computing problems (physics, biology, math)
GPT-6 Astra54.2
MiniMax M3.1 Flash53.4
DeepSWE 1.1 DeepSWE 1.1: Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition
GPT-6 Astra72.8
MiniMax M3.1 Flash—
SWE-bench Pro (v2 full) SWE-bench Pro (v2 full): Agents resolve repository issues under the network-locked v2 protocol; full and 51-task Hard results are separate
GPT-6 Astra · high96.9
MiniMax M3.1 Flash—
SWE-bench Pro (v2 hard) SWE-bench Pro (v2 hard): Agents resolve repository issues under the network-locked v2 protocol; full and 51-task Hard results are separate
GPT-6 Astra · high90.2
MiniMax M3.1 Flash—
Terminal-Bench Science 0.1 Terminal-Bench Science 0.1: Agents complete expert-authored research workflows across five scientific domains in a terminal
GPT-6 Astra · max63.3
MiniMax M3.1 Flash—
MultiLangBenchlab MultiLangBench: Models build a SQLite-compatible database engine in 20 programming languages
GPT-6 Astra87.4
MiniMax M3.1 Flash—
MirrorCode MirrorCode: Agents reimplement software projects by observing their behavior
GPT-6 Astra · high46.7
MiniMax M3.1 Flash—
| Benchmark | GPT-6 Astra | MiniMax M3.1 Flash |
|---|---|---|
| Terminal-Bench 4.0 | 49.5 | 25.8 |
| FrontierCode 1.1 | 47 · max | 38.6 |
| WebDev Arena | 1,774 · max | 1,587 |
| SciCode | 54.2 | 53.4 |
| DeepSWE 1.1 | 72.8 | — |
| SWE-bench Pro (v2 full) | 96.9 · high | — |
| SWE-bench Pro (v2 hard) | 90.2 · high | — |
| Terminal-Bench Science 0.1 | 63.3 · max | — |
| MultiLangBenchlab | 87.4 | — |
| MirrorCode | 46.7 · high | — |
Specs
| Spec | GPT-6 Astra | MiniMax M3.1 Flash |
|---|---|---|
| Input, per 1M | $10 | $0.15 |
| Output, per 1M | $50 | $0.60 |
| Cost per task? | $1.54 | $0.22 |
| Speed, tok/s | 59 | 105 |
| Context | 1.05M | 1M |
| Weights | Closed | Closed |
| Released | Sep 2026 | Oct 2026 |
| See images | ✓ | ✓ |
| Read PDFs | Partial | — |
| Hear audio | — | — |
| Watch video | — | ✓ |
| Call tools | ✓ | Partial |
| Force a tool call | Partial | ? |
| Return strict JSON | ✓ | Partial |
| Search the web | ✓ | ? |
| Run code | ✓ | ? |
| Use a computer | ✓ | ? |
What breaks when you switch
GPT-6 Astra → MiniMax M3.1 Flash
Context: 1.05M → 1M
Complete model response must be appended›
In multi-turn tools, preserve the complete assistant response, including reasoning_content and tool_calls, to maintain reasoning continuity.
MiniMax M3.1 Flash → GPT-6 Astra
Monitoring can return HTTP 403›
Misalignment monitoring can return HTTP 403 before streaming or errors after output; match misalignment_policy_violation and handle stream errors.
No none reasoning effort›
The none reasoning effort is unsupported on GPT-6 Astra and GPT-6.1 Sol; use low instead.
Remove unsupported sampling parameters›
When reasoning effort is not none, remove temperature, top_p and top_logprobs, plus Chat Completions logprobs or Responses output logprobs inclusion.