Compare / head-to-head
GPT-6 Astravs
Qwen3 4B Instruct 2507
GPT-6 Astra leads 3 of 3 shared benchmarks. GPT-6 Astra has the larger context window (1.1M tokens).
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
3 – 0
GPT-6 Astra leads
Cheaper per token
—
list price, input + output
Larger context
GPT-6 Astra
1.1M tokens
Providers
2 providers
OpenAI · Qwen
GPT-6 Astra
OpenAI · released 2026-09-03
textvision
Qwen3 4B Instruct 2507
Qwen · released 2025-08-06
text
- Context
- 262K
- Max out
- —
- Input /1M
- —
- Output /1M
- —
- Cached /1M
- —
- Scores
- 18 · 7 core
Quality
Benchmark matrix
3 shared · 30 only GPT-6 Astra · 15 only Qwen3 4B Instruct 2507| Benchmark | GPT-6 Astra | Qwen3 4B Instruct 2507 | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 3 | ||||
| GPQA Diamond | 96% ↗ | 45.8% ↗epoch run | +50.2 pt | GPT-6 Astra |
| Chess Puzzles (Epoch AI run) | 72% ↗epoch run | 4% ↗epoch run | +68 pt | GPT-6 Astra |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | 100% ↗epoch run | 52.2% ↗epoch run | +47.8 pt | GPT-6 Astra |
| Only GPT-6 Astra reports · 30 | ||||
| LMArena Elo | 1477.1 ↗ | not reported | — | — |
| APEX-Agents (Mercor) | 64.7% ↗ | not reported | — | — |
| AutomationBench 1.0.6 | 41.4% ↗ | not reported | — | — |
| BALROG (BALROG) | 68.3% ↗ | not reported | — | — |
| DeepSWE v1.1 (Datacurve) | 74.1% ↗ | not reported | — | — |
| EBR-bench (Epoch AI run) | 76.2% ↗epoch run | not reported | — | — |
| FrontierMath Tier 4 v2 (Epoch AI run) | 97.6% ↗epoch run | not reported | — | — |
| FrontierMath Tiers 1-3 v2 (Epoch AI run) | 93.7% ↗epoch run | not reported | — | — |
| FrontierSWE V2 (Proximal Labs) | 65.5% ↗ | not reported | — | — |
| Furniture Assembly (Epoch AI run) | 80% ↗epoch run | not reported | — | — |
| GSO Opt@1 (GSO) | 77.5% ↗ | not reported | — | — |
| LiveBench Agentic Coding (LiveBench) | 57.3% ↗ | not reported | — | — |
| LiveBench Coding (LiveBench) | 80.4% ↗ | not reported | — | — |
| LiveBench Data Analysis (LiveBench) | 83% ↗ | not reported | — | — |
| LiveBench Instruction Following (LiveBench) | 75.6% ↗ | not reported | — | — |
| LiveBench Language (LiveBench) | 89.4% ↗ | not reported | — | — |
| LiveBench Mathematics (LiveBench) | 96.8% ↗ | not reported | — | — |
| LiveBench Reasoning (LiveBench) | 92.7% ↗ | not reported | — | — |
| LMArena Agent (LMArena) | 0.1227 ↗ | not reported | — | — |
| LMArena Vision (LMArena) | 1284.3 ↗ | not reported | — | — |
| LMArena WebDev (LMArena) | 1787.7 ↗ | not reported | — | — |
| MirrorCode (Epoch AI run) | 46.7% ↗epoch run | not reported | — | — |
| Mystery Game Puzzles (Epoch AI run) | 84% ↗epoch run | not reported | — | — |
| SAGE (Vals AI) | 46.4% ↗ | not reported | — | — |
| SimpleBench (SimpleBench) | 86.5% ↗ | not reported | — | — |
| SimpleQA Verified | 75.6% ↗epoch run | not reported | — | — |
| Terminal-Bench 4.0 (Vals AI) | 59.6% ↗ | not reported | — | — |
| Terminal-Bench Science 0.1 | 64.6% ↗ | not reported | — | — |
| Vending-Bench 2 (Andon Labs) | 15514.7 ↗ | not reported | — | — |
| WeirdML (Håvard Tveit Ihle) | 93.6% ↗ | not reported | — | — |
| Only Qwen3 4B Instruct 2507 reports · 15 | ||||
| MMLU-Pro | not reported | 69.6% ↗ | — | — |
| Aider-Polyglot | not reported | 12.9% ↗ | — | — |
| AIME25 | not reported | 47.4% ↗ | — | — |
| Arena-Hard v2 | not reported | 43.4% ↗ | — | — |
| BFCL-v3 | not reported | 61.9% ↗ | — | — |
| GPQA | not reported | 62% ↗ | — | — |
| HMMT25 | not reported | 31% ↗ | — | — |
| IFEval | not reported | 83.4% ↗ | — | — |
| LiveBench 20241125 | not reported | 63% ↗ | — | — |
| LiveCodeBench v6 (25.02-25.05) | not reported | 35.1% ↗ | — | — |
| MMLU-Redux | not reported | 84.2% ↗ | — | — |
| MultiPL-E | not reported | 76.8% ↗ | — | — |
| SuperGPQA | not reported | 42.8% ↗ | — | — |
| TAU2-Retail | not reported | 40.4% ↗ | — | — |
| ZebraLogic | not reported | 80.2% ↗ | — | — |
Scores tagged "epoch" or "matharena" are independent runs, used only where the lab has not published its own; ⚠ marks rows MathArena flags as released after the competition. Higher is better on every row. Δ is GPT-6 Astra minus Qwen3 4B Instruct 2507 in the benchmark's own unit. "Not reported" means the lab has not published that figure; it is not a zero. ↗ opens the source.
Specs & pricing
Side by side
Official pricing pages and model cards| Spec | GPT-6 Astra | Qwen3 4B Instruct 2507 | Edge |
|---|---|---|---|
| Context window | 1.1M tokens | 262K tokens | GPT-6 Astra |
| Max output | 128K tokens | — | — |
| Input price / 1M | $10 ↗ | — | — |
| Output price / 1M | $50 ↗ | — | — |
| Cached input / 1M | $1 ↗ | — | — |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | $60 | — | — |
| Modalities | text · vision | text | GPT-6 Astra |
| Released | 2026-09-03 | 2025-08-06 | — |
| Cited benchmark scores | 39 | 18 | — |
Reliability
Provider status
Live from /statusMore matchups
GPT-6 Astra vs …
Models sharing the most benchmarksMore matchups
Qwen3 4B Instruct 2507 vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run GPT-6 Astra and Qwen3 4B Instruct 2507 on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.