Quality
Known Good Index and the capability indices, side by side
Each index is min-max normalised across tracked models and scaled 0–100 from published evaluations. We do not run these evaluations.
| Metric | Microsoft: Phi 4 | OpenAI: gpt-oss-120b |
|---|---|---|
| Known Good Index | 33 | 61 |
| Mathematics Index | 39 | 89 |
| Science Index | 53 | 77 |
| Reasoning Index | 47 | — |
| Agentic Index | — | 18 |
| Openness Index | 60 | 60 |
Quality Breakdown
Every evaluation both models report a published score for
Bar = Microsoft: Phi 4 · marker = OpenAI: gpt-oss-120b · scores ingested from Epoch AI (CC-BY 4.0)
Right-hand figures read Microsoft: Phi 4 / OpenAI: gpt-oss-120b. Microsoft: Phi 4 leads on 0 of 2, OpenAI: gpt-oss-120b on 2.
Arena Elo
Human preference rating from blind pairwise votes
Elo from blind pairwise votes, ingested from Arena. Higher is better.
| Metric | Microsoft: Phi 4 | OpenAI: gpt-oss-120b |
|---|---|---|
| Text Arena | 1,217 | 1,366 |
| Text Arena (style controlled) | 1,256 | 1,352 |
Price & Cost
Published API pricing by token type
USD per 1M tokens. Blended is a 3:1 input:output mix. Live from OpenRouter, cross-checked against models.dev. Lower is better.
| Metric | Microsoft: Phi 4 | OpenAI: gpt-oss-120b |
|---|---|---|
| Input $/1M | $0.07 | $0.04 |
| Output $/1M | $0.14 | $0.17 |
| Blended $/1M | $0.09 | $0.07 |
| Cache read $/1M | — | $0.015 |
Context Window
Maximum input accepted and maximum output emitted
Provider-declared limits, from metadata via models.dev and OpenRouter. Higher is better.
| Metric | Microsoft: Phi 4 | OpenAI: gpt-oss-120b |
|---|---|---|
| Context window | 16k | 131k |
| Max output tokens | 16k | 131k |
Speed & Latency
Median throughput and time to first token across hosting providers
Medians across all tracked providers, derived from OpenRouter endpoint telemetry.
| Metric | Microsoft: Phi 4 | OpenAI: gpt-oss-120b |
|---|---|---|
| Output tokens/s | 55 | 173 |
| Time to first token | 0.2s | 0.7s |
Providers
Which hosts serve each model, and at what price and speed
| Provider | Microsoft: Phi 4 blended | Microsoft: Phi 4 tok/s | OpenAI: gpt-oss-120b blended | OpenAI: gpt-oss-120b tok/s |
|---|---|---|---|---|
| DeepInfra | $0.09 | 38 | $0.07 | 282 |
| Cerebras | — | — | $0.45 | 505 |
| Groq | — | — | $0.26 | 445 |
| SambaNova | — | — | $0.34 | 304 |
| BaseTen | — | — | $0.20 | 255 |
| Nebius | — | — | $0.26 | 255 |
| Mara | — | — | $0.30 | 198 |
| Parasail | — | — | $0.26 | 198 |
| Google · first party | — | — | $0.16 | 197 |
| Together | — | — | $0.26 | 108 |
| Amazon Bedrock · first party | — | — | $0.26 | 103 |
| Phala | — | — | $0.26 | 86 |
| Novita | — | — | $0.10 | 67 |
| DigitalOcean | — | — | $0.17 | 63 |
| Mancer 2 | — | — | $0.17 | 49 |
| SiliconFlow | — | — | $0.15 | 29 |
| CoreWeave | — | — | $0.07 | 24 |
Per-host pricing and telemetry from OpenRouter. A dash means that host does not serve that model.