Claude Opus 5.5 vs GPT-6 Astra: Benchmark Comparison
Claude Opus 5.5 has the higher score on 15 of 24 shared benchmarks; GPT-6 Astra leads on 8.
The largest observed score gap is 23.28 pts on SRE Bench , where GPT-6 Astra leads.
Reported ±1 standard-error ranges overlap on 12 of 22 shared benchmarks with comparable uncertainty data. This is not a pairwise statistical significance test.
Shared benchmark results
Scores and reported standard errors come from the latest visible Vals benchmark version available for both models.
| Benchmark | Claude Opus 5.5 | GPT-6 Astra | Gap | Reported uncertainty |
|---|---|---|---|---|
| Vals Index | 66.97% ±0.89 | 63.13% ±1.19 | 3.85 pts | Reported ±1 SE ranges do not overlap |
| Vals RSI Index | 37.31% | 27.07% | 10.24 pts | Uncertainty comparison unavailable |
| Harvey's Legal Agent Benchmark | 3.75% ±1.47 | 5.42% ±1.17 | 1.67 pts | Reported ±1 SE ranges overlap |
| Legal Research Bench | 50.48% ±3.48 | 39.42% ±3.40 | 11.06 pts | Reported ±1 SE ranges do not overlap |
| EMB | 75.94% ±2.38 | 71.70% ±2.48 | 4.24 pts | Reported ±1 SE ranges overlap |
| Finance Agent (v2) | 58.59% ±0.17 | 53.54% ±2.08 | 5.05 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 70.50% ±3.15 | 63.34% ±3.15 | 7.16 pts | Reported ±1 SE ranges do not overlap |
| MedCode | 49.80% ±2.27 | 48.49% ±2.13 | 1.31 pts | Reported ±1 SE ranges overlap |
| MedScribe | 91.43% ±1.93 | 87.91% ±1.94 | 3.52 pts | Reported ±1 SE ranges overlap |
| ProofBench v1.1 | 100.00% ±0.00 | 99.00% ±1.00 | 1.00 pts | Reported ±1 SE ranges overlap |
| BioMysteryBench | 79.26% ±0.98 | 79.26% ±0.98 | 0.00 pts | Reported ±1 SE ranges overlap |
| MysteryMechanism | 49.55% ±3.36 | 53.15% ±3.36 | 3.60 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench Science | 47.14% ±6.01 | 62.86% ±5.82 | 15.71 pts | Reported ±1 SE ranges do not overlap |
| SAGE | 45.83% ±3.36 | 46.37% ±3.44 | 0.54 pts | Reported ±1 SE ranges overlap |
| Code Migration | 66.65% ±4.33 | 67.74% ±4.22 | 1.09 pts | Reported ±1 SE ranges overlap |
| IOI | 95.06% ±4.94 | 100.00% ±0.00 | 4.94 pts | Reported ±1 SE ranges overlap |
| ProgramBench | 18.50% ±2.75 | 5.50% ±1.62 | 13.00 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench 4.0 | 65.15% ±0.00 | 59.60% ±4.40 | 5.56 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench 1-100 | 30.36% ±4.83 | 27.64% ±4.08 | 2.72 pts | Reported ±1 SE ranges overlap |
| Vibe Code Bench v1.1 | 90.29% ±1.53 | 89.59% ±2.17 | 0.70 pts | Reported ±1 SE ranges overlap |
| CyberBench v1.1 | 55.36% ±5.13 | 41.07% ±2.56 | 14.28 pts | Reported ±1 SE ranges do not overlap |
| SRE Bench | 33.59% ±2.92 | 56.87% ±3.07 | 23.28 pts | Reported ±1 SE ranges do not overlap |
| CUA-bench | 14.00% | 19.17% | 5.17 pts | Uncertainty comparison unavailable |
| Time Horizon Index: KSP | 91.33% ±0.00 | 90.50% ±0.00 | 0.83 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | Claude Opus 5.5 average | GPT-6 Astra average |
|---|---|---|
| Legal | 27.12% | 22.42% |
| Finance | 68.34% | 62.86% |
| Healthcare | 70.61% | 68.20% |
| Math | 100.00% | 99.00% |
| Science | 58.65% | 65.09% |
| Education | 45.83% | 46.37% |
| Coding | 61.00% | 58.35% |
| Cyber | 44.47% | 48.97% |
Cost and latency
Cost per test appears only where the benchmark reports it for both models. Latency is the measured completion time for that benchmark.
| Benchmark | Claude Opus 5.5 cost | GPT-6 Astra cost | Claude Opus 5.5 latency | GPT-6 Astra latency |
|---|---|---|---|---|
| Vals Index | $32.14 | $18.46 | 1h19m | 27m48s |
| Vals RSI Index | $411.86 | $759.51 | 90h00m | 90h00m |
| Harvey's Legal Agent Benchmark | $21.38 | $26.16 | 49m00s | 24m49s |
| Legal Research Bench | $25.16 | $10.58 | 1h52m | 24m51s |
| EMB | $11.02 | $5.83 | 35m06s | 13m24s |
| Finance Agent (v2) | $9.22 | $6.82 | 33m08s | 11m33s |
| Tax Agent Bench | $15.09 | $5.80 | 1h07m | 15m01s |
| MedCode | N/A | N/A | 4m07s | 102.86s |
| MedScribe | N/A | N/A | 7m22s | 2m34s |
| ProofBench v1.1 | $0.96 | $1.71 | 5m51s | 5m08s |
| BioMysteryBench | $3.64 | $1.72 | 12m19s | 4m46s |
| MysteryMechanism | $4.62 | $1.56 | 19m38s | 8m09s |
| Terminal-Bench Science | $19.12 | $20.80 | 2h28m | 1h49m |
| SAGE | N/A | N/A | 114.20s | 41.63s |
| Code Migration | $112.97 | $44.36 | 3h19m | 58m30s |
| IOI | $5.26 | $6.50 | 17m15s | 31m40s |
| ProgramBench | $68.05 | $11.57 | 2h24m | 27m02s |
| Terminal-Bench 4.0 | $13.20 | $9.58 | 1h04m | 35m42s |
| Vibe Code Bench 1-100 | $44.12 | $84.72 | 2h28m | 3h26m |
| Vibe Code Bench v1.1 | $57.92 | $38.51 | 1h36m | 39m04s |
| CyberBench v1.1 | $3.48 | $1.84 | 18m11s | 4m10s |
| SRE Bench | $34.65 | $10.24 | 1h37m | 17m51s |
| CUA-bench | $188.04 | $1821.37 | 14.71s | 32.09s |
| Time Horizon Index: KSP | $1303.37 | $4206.59 | N/A | N/A |
Results available only for Claude Opus 5.5
- Public Benefits Bench v1.1
Results available only for GPT-6 Astra
None.