Comparison · Anthropic vs OpenAI
Claude Opus 5.5 vs GPT-6 Astra: pricing, allowances and benchmarks
Claude Opus 5.5 (Anthropic) and GPT-6 Astra (OpenAI): 2 shared subscriptions and 6 shared leaderboard snapshots, with workloads, sample dates and benchmark configurations.
Plan & benchmark data snapshot · How real prices are computed
As of 2026-10-16, 1 of 2 shared subscriptions use matching normalized workloads: Claude Opus 5.5 has a lower recorded token cost on 1, GPT-6 Astra on 0, with 0 tied. Their highest archived configurations score 57.62 and 52.67 on Artificial Analysis Intelligence Index v4.3; all 6 shared leaderboard snapshots, configurations and sources appear below.
- Shared subscriptions
- 2Both models available
- Comparable workloads
- 1Used for price ratios
- Shared leaderboards
- 6Highest archived configurations
- Official API prices
- 2/ 2Models with list prices
Plot both on the price–capability chart
Same plan, side by side
Each row is a subscription that includes both models at the same fee. Real cost = monthly fee ÷ the recorded monthly allowance. Each model's allowance assumes using the plan for that model; the columns cannot be added together. Click a price for its calculation and sources. Ratios use only matching normalized workloads.
| Plan / provider | Monthly fee | Claude Opus 5.5 | GPT-6 Astra | Recorded token cost |
|---|---|---|---|---|
| Devin MaxDevin | $200/mo | $0.0299 /MTok 6.69B tokens/month Anthropic workload 2026-09-23 · sampled · confidence Medium | $0.1794 /MTok 1.115B tokens/month Standard workload 2026-09-14 · sampled · confidence Medium | Different workloads |
| Droid MaxFactory | $200/mo | $0.07918 /MTok 2.526B tokens/month Anthropic workload 2026-09-29 · sampled · confidence Medium | $0.1979 /MTok 1.01B tokens/month Anthropic workload 2026-09-29 · derived; uses the data date of Droid Max · claude-opus-5.5 · confidence Medium | Claude Opus 5.5 lower2.5× price ratio (higher/lower) |
Leaderboards head to head
Each model's highest archived configuration on each shared leaderboard snapshot, ranked among the models archived in this dataset. Harness, reasoning effort and service mode can differ; self-reported and estimated results are marked. These reference configurations are not measurements of the subscriptions above. Overlapping leaderboards do not form an overall ability rating.
| Leaderboard | Metric | Claude Opus 5.5 | GPT-6 Astra | Higher archived score |
|---|---|---|---|---|
| Artificial Analysis Intelligence Index v4.3 | Intelligence Index | 57.62 #1 / 60 · Claude Opus 5.5 (max with fallback) Max 2026-10-01 · Record | 52.67 #4 / 60 · GPT-6 Astra (max) Max 2026-10-01 · Record | Claude Opus 5.5+4.95 |
| Terminal-Bench 4.0 | Resolution Rate % | 66.4 self-reported #1 / 18 · Claude Code - Claude Opus 5.5 (default) - self-reported Claude Code 2026-09-23 · Record | 58.18 #2 / 18 · Codex - GPT-6 Astra (max) Codex · Max Reported interval 55.39–60.97 2026-10-01 · Record | Claude Opus 5.5+8.22 |
| Terminal-Bench 4.0 (AA) | Resolution Rate % | 59.6 #2 / 53 · Artificial Analysis - Claude Opus 5.5 (max with fallback) Artificial Analysis · Max 2026-10-01 · Record | 59.6 #2 / 53 · Artificial Analysis - GPT-6 Astra (xhigh) Artificial Analysis · xhigh 2026-10-01 · Record | Equal score |
| Code Arena | WebDev Overall | Arena Score | 1,818 #1 / 54 · claude-opus-5.5-max Max Reported interval 1,801.5–1,834.5 2026-10-01 · Record | 1,789 #2 / 54 · gpt-6-astra-max Max Reported interval 1,778.6–1,799.4 2026-10-01 · Record | Claude Opus 5.5+29 |
| Agent Arena · Overall (Ranking / Pareto) | Net Improvement % | 13.78 #2 / 39 · Claude Opus 5.5 (High) High Reported interval 11.54–16.02 2026-10-01 · Record | 12.18 #3 / 39 · GPT 6 Astra (Max) Max Reported interval 9.9–14.46 2026-10-01 · Record | Claude Opus 5.5+1.6 |
| Artificial Analysis Coding Agent Index v1.5 | Coding Agent Index | 65.99 #2 / 19 · Claude Code - Opus 5.5 (max) Claude Code · Max 2026-10-01 · Record | 61.65 #5 / 19 · Codex - GPT-6 Astra (max) Codex · Max 2026-10-01 · Record | Claude Opus 5.5+4.34 |
Official API pricing
Each row uses its listed pricing conditions. Cache writes are shown separately; subscription costs above retain their recorded quota basis. An em dash means that rate is not recorded.
| Model / pricing conditions | Cached input | Input | Cache write | Output |
|---|---|---|---|---|
| Claude Opus 5.5Upstream referenceUpstream price reference | $0.2 | $4 | — | $20 |
| GPT-6 AstraStandard · Global · ≤ 272,000 input tokensOfficial source · Checked All pricing tiers | $1 | $10 | $12.5 | $50 |
How to read this comparison
Price ratios compare recorded USD/MTok under the same known workload. Raw measured totals, different workloads and unmetered access remain visible but are excluded from the lower-cost counts. Recorded allowances assume using the quota; your effective cost depends on utilization, cache mix, output mix and plan limits. Qualitative confidence is not an error bar.
- Anthropic workload: 97% cache reads / 2.5% cache writes / 0.5% output
- Standard workload: 97% cache reads / 2.5% input / 0.5% output