Arena
Head-to-head under the same evidence
Pick two live endpoints. Scores, speed, cost, and agent efficiency stay in separate rows. Nothing is averaged into a vanity total.
If you optimize for effective work rate, take Grok 4 on xAI. If you optimize for agent task completion, take GPT-5. Dimensions won: Grok 4 6 · GPT-5 4.
xAI
Grok 4
xAI API
Universe
87.3
EWR
0.925
TTFT
410 ms
tok/s
138
LIVE_MEASURED
OpenAI
GPT-5
OpenAI API
Universe
88.6
EWR
0.268
TTFT
520 ms
tok/s
95
LIVE_MEASURED
| Dimension | Grok 4 | GPT-5 | Edge |
|---|---|---|---|
| Universe index | 87.3 | 88.6 | GPT-5 |
| Coding | 81.2 | 85.4 | GPT-5 |
| Agents | 67.5 | 70.1 | GPT-5 |
| Science | 90.1 | 88.4 | Grok 4 |
| Effective work rate | 0.925 | 0.268 | Grok 4 |
| $ / correct task | $0.095 | $0.262 | Grok 4 |
| TTFT p50 | 410 ms | 520 ms | Grok 4 |
| Output tok/s | 138 | 95 | Grok 4 |
| Availability | 99.94% | 99.91% | Grok 4 |
| Useful action efficiency | 0.039 | 0.042 | GPT-5 |