MicrosoftPhi-4
Best configuration ranks #523 of 610 on the BenchLeader Index at 39.7 ±4.2. Last measured 2 Sept 2026. Released 11 Dec 2024.
- Blended price
- $0.219/M
- $0.125 in · $0.500 out
- Output speed
- 41 tok/s
- First answer
- 2.54 s
- first token 0.20 s
- Context
- 16k
- Overall index40
- Reasoning42
- Coding42
- Agents & tools40
- Maths42
- Knowledge38
- Instruction following28
- Human preference41
- Long context25
- Composite36
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Default” means the publisher did not say which setting was used.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| thinking | – | – | 41 tok/s | 2.54 s | $0.0002 | |
| defaultbest | 39.7 | #523 | 41 tok/s | 2.54 s | $0.0002 | Agents & tools 40 · Coding 42 · Composite 36 · Human preference 41 · Instruction following 28 · Knowledge 38 · Long context 25 · Maths 42 · Reasoning 42 |
Benchmark results
One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | default | Source |
|---|---|---|
| GPQA Diamond | 56.1%#186 | Epoch AI Benchmarking Hub |
| LMArena Hard Prompts | 1278#253 | LMArena |
| GPQA Diamond (AA)not in index | 57.5%#372 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 3.8%#483 | Artificial Analysis |
Coding
| Benchmark | default | Source |
|---|---|---|
| LMArena Coding | 1306#250 | LMArena |
Agents & tools
| Benchmark | default | Source |
|---|---|---|
| Terminal-Bench Hard | 3.8%#283 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 0.0%#375 | Artificial Analysis |
| BFCL Overall | 28.8%#52 | Berkeley Function Calling Leaderboard |
Maths
| Benchmark | default | Source |
|---|---|---|
| OTIS Mock AIME | 13.8%#208 | Epoch AI Benchmarking Hub |
| MATH Level 5 | 64.9%#41 | Epoch AI Benchmarking Hub |
Knowledge
| Benchmark | default | Source |
|---|---|---|
| AA-Omniscience | -55.7#365 | Artificial Analysis |
Instruction following
| Benchmark | default | Source |
|---|---|---|
| IFBench | 23.5%#378 | Artificial Analysis |
Human preference
| Benchmark | default | Source |
|---|---|---|
| LMArena Text | 1256#264 | LMArena |
Long context
| Benchmark | default | Source |
|---|---|---|
| AA-LCR | 0.0%#419 | Artificial Analysis |
Composite
| Benchmark | default | Source |
|---|---|---|
| Epoch Capabilities Indexnot in index | 130.4#131 | Epoch AI Benchmarking Hub |
| AA Intelligence Index | 5.9#464 | Artificial Analysis |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| DeepInfra | 48 tok/s | 0.20 s | $0.070 | $0.140 | 16k | bf16 |
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0002 | – | 9.8 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0018 | – | 17.1 s |
| Code edit | 6,000 / 1,500 | $0.0015 | – | 39.0 s |
| Agentic coding session | 60,000 / 4,000 | $0.0095 | – | 1.7 min |
| Structured extraction | 2,000 / 200 | $0.0004 | – | 7.4 s |
See also
Data as of 9 Sept 2026. Compare these configurations.