Compare / head-to-head
Phi-4vs
Qwen3 235B-A22B Thinking 2507
Qwen3 235B-A22B Thinking 2507 leads 6 of 6 shared benchmarks. Qwen3 235B-A22B Thinking 2507 has the larger context window (262K tokens).
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
0 – 6
Qwen3 235B-A22B Thinking 2507 leads
Cheaper per token
—
list price, input + output
Larger context
Qwen3 235B-A22B Thinking 2507
262K tokens
Providers
2 providers
Microsoft · Qwen
Phi-4
Microsoft · released 2024-12-12
text
- Context
- 16K
- Max out
- 16K
- Input /1M
- —
- Output /1M
- —
- Cached /1M
- —
- Scores
- 17 · 7 core
Qwen3 235B-A22B Thinking 2507
Qwen · released 2025-07-25
text
Quality
Benchmark matrix
6 shared · 7 only Phi-4 · 17 only Qwen3 235B-A22B Thinking 2507| Benchmark | Phi-4 | Qwen3 235B-A22B Thinking 2507 | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 6 | ||||
| GPQA Diamond | 56.1% ↗ | 80.1% ↗epoch run | -24 pt | Qwen3 235B-A22B Thinking 2507 |
| MMLU-Pro | 70.4% ↗ | 84.4% ↗ | -14 pt | Qwen3 235B-A22B Thinking 2507 |
| LMArena Elo | 1216.6 ↗ | 1400 ↗ | -183.4 | Qwen3 235B-A22B Thinking 2507 |
| Chess Puzzles (Epoch AI run) | 1% ↗epoch run | 12% ↗epoch run | -11 pt | Qwen3 235B-A22B Thinking 2507 |
| IFEval | 63% ↗ | 87.8% ↗ | -24.8 pt | Qwen3 235B-A22B Thinking 2507 |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | 13.8% ↗epoch run | 86.7% ↗epoch run | -72.9 pt | Qwen3 235B-A22B Thinking 2507 |
| Only Phi-4 reports · 7 | ||||
| MMLU | 84.8% ↗ | not reported | — | — |
| HumanEval | 82.6% ↗ | not reported | — | — |
| ArenaHard | 75.4% ↗ | not reported | — | — |
| BALROG (BALROG) | 11.6% ↗ | not reported | — | — |
| HumanEval+ | 82.8% ↗ | not reported | — | — |
| MATH | 80.4% ↗ | not reported | — | — |
| MATH Level 5 (Epoch AI run) | 64.9% ↗epoch run | not reported | — | — |
| Only Qwen3 235B-A22B Thinking 2507 reports · 17 | ||||
| AIME25 | not reported | 92.3% ↗ | — | — |
| Arena-Hard v2 | not reported | 79.7% ↗ | — | — |
| BFCL-v3 | not reported | 71.9% ↗ | — | — |
| GPQA | not reported | 81.1% ↗ | — | — |
| HLE (text-only) | not reported | 18.2% ↗ | — | — |
| HMMT25 | not reported | 83.9% ↗ | — | — |
| LiveBench 20241125 | not reported | 78.4% ↗ | — | — |
| LiveCodeBench v6 (25.02-25.05) | not reported | 74.1% ↗ | — | — |
| MMLU-Redux | not reported | 93.8% ↗ | — | — |
| Mystery Game Puzzles (Epoch AI run) | not reported | 9% ↗epoch run | — | — |
| OJBench | not reported | 32.5% ↗ | — | — |
| SimpleQA Verified | not reported | 40.4% ↗epoch run | — | — |
| SuperGPQA | not reported | 64.9% ↗ | — | — |
| TAU2-Airline | not reported | 58% ↗ | — | — |
| TAU2-Retail | not reported | 71.9% ↗ | — | — |
| Vending-Bench 2 (Andon Labs) | not reported | -11.34 ↗ | — | — |
| WeirdML (Håvard Tveit Ihle) | not reported | 41% ↗ | — | — |
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 Phi-4 minus Qwen3 235B-A22B Thinking 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 | Phi-4 | Qwen3 235B-A22B Thinking 2507 | Edge |
|---|---|---|---|
| Context window | 16K tokens | 262K tokens | Qwen3 235B-A22B Thinking 2507 |
| Max output | 16K tokens | — | — |
| Input price / 1M | — | $0.23 ↗ | — |
| Output price / 1M | — | $2.3 ↗ | — |
| Cached input / 1M | — | — | — |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | — | $2.53 | — |
| Modalities | text | text | Tie |
| Released | 2024-12-12 | 2025-07-25 | — |
| Cited benchmark scores | 17 | 23 | — |
Reliability
Provider status
Live from /statusMore matchups
Phi-4 vs …
Models sharing the most benchmarksMore matchups
Qwen3 235B-A22B Thinking 2507 vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run Phi-4 and Qwen3 235B-A22B Thinking 2507 on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.