Model comparison
Claude Opus 5.5 vs GPT-6 Astra
Anthropic against OpenAI, compared on context, price, and verified benchmark results.
Catalog record checked September 26, 2026Individual provider fields may changeEvidence confidence: High — see receipts below
Anthropic
Claude Opus 5.5
Frontier
OpenAI
GPT-6 Astra
Frontier
| Specification | Claude Opus 5.5 | GPT-6 Astra |
|---|---|---|
| Provider | ||
| Provider | Anthropic | OpenAI |
| Tier | ||
| Tier | Frontier | Frontier |
| Context window | ||
| Context window | 1M | Winner: 1.05M |
| Max output | ||
| Max output | 128K | 128K |
| Input / 1M tokens | ||
| Input / 1M tokens | Winner: $4 | $10 |
| Output / 1M tokens | ||
| Output / 1M tokens | Winner: $20 | $50 |
| Weights | ||
| Weights | Closed | Closed |
| Parameters | ||
| Parameters | Not disclosedUnverified | overall flagship (reasoning, coding, computer use) |
| Reasoning levels | ||
| Reasoning levels | adaptive | low, medium, high, xhigh, max |
| Modalities | ||
| Modalities | text, image | text, image |
| API model id | ||
| API model id | claude-opus-5-5 | gpt-6-astra |
| Released | ||
| Released | September 22, 2026 | September 3, 2026 |
| Artificial Analysis Intelligence Index [max] (2026-09-26) | ||
| Artificial Analysis Intelligence Index [max] (2026-09-26) | Winner: 57.6 | 52.7 |
| Humanity's Last Exam [max] (2026-09-26) | ||
| Humanity's Last Exam [max] (2026-09-26) | 61.4 | Not verifiedUnverified |
| Humanity's Last Exam (2026-09-03) | ||
| Humanity's Last Exam (2026-09-03) | Not verifiedUnverified | 57.2 |
| Terminal-Bench 4.0 [xhigh] (2026-09-22) | ||
| Terminal-Bench 4.0 [xhigh] (2026-09-22) | Winner: 66.4 | 59.6 |
| DeepSWE 1.1 (2026-09-03) | ||
| DeepSWE 1.1 (2026-09-03) | Not verifiedUnverified | 74.1 |
| OSWorld 2.0 (offline) (2026-09-03) | ||
| OSWorld 2.0 (offline) (2026-09-03) | Not verifiedUnverified | 72.6 |
| GPQA Diamond (2026-09-03) | ||
| GPQA Diamond (2026-09-03) | Not verifiedUnverified | 96 |
| Terminal-Bench 2.1 [xhigh] (2026-09-26) | ||
| Terminal-Bench 2.1 [xhigh] (2026-09-26) | Not verifiedUnverified | 89.1 |
Prices are USD per million tokens at standard rates, excluding batch and caching discounts. Bold indicates the better figure where one is objectively better. Values we could not confirm from the provider are shown as “Not verified” rather than estimated.
Benchmark receipts
Where each score comes from, and how far it can be compared across models.
- 2026-09-26Artificial Analysis Terminal-Bench 2.1 (independent, Terminus 2 harness)
Agentic terminal work: multi-step tasks executed in a sandboxed shell environment.
Comparable with caveatNot comparable with Terminal-Bench 3.0 or 4.0 (different task sets) or v1; harness configuration (container, time limits) also shifts results.
- 2026-09-26Artificial Analysis Terminal-Bench 4.0 (independent AA run, part of Intelligence Index v4.3.2)
Agentic terminal work: long multi-step tasks executed in a sandboxed shell environment.
Comparable with caveatNOT comparable with Terminal-Bench 2.x or 3.0 — 4.0 uses a new, non-overlapping task set. Within 4.0, scores from different harnesses (tbench.ai agent entries vs Artificial Analysis runs) are not interchangeable.
- 2026-09-26Artificial Analysis Humanity's Last Exam (independent, 2,158 text-only questions)
Frontier-knowledge ceiling: extremely hard multi-domain questions written to be near-impossible without deep expertise.
Comparable with caveatSubset/tool configurations (text-only vs with-tools) differ between vendors and materially change scores; must match configuration to compare.
- 2026-09-26Artificial Analysis
Composite index blending reasoning, knowledge, and coding evals into one 0–100 score.
Comparable with caveatOnly same-version scores are comparable. v4.3.2 was rebaselined (the top score fell from 66 on v4.1.1 to ~58), so v4.1.1 figures must not be compared with v4.3.2 figures — check measuredAt.
- 2026-09-22Anthropic — Claude Opus 5.5 announcement (vendor; conflicts with AA's independent 59.6 — harness difference preserved)
Agentic terminal work: long multi-step tasks executed in a sandboxed shell environment.
Comparable with caveatNOT comparable with Terminal-Bench 2.x or 3.0 — 4.0 uses a new, non-overlapping task set. Within 4.0, scores from different harnesses (tbench.ai agent entries vs Artificial Analysis runs) are not interchangeable.
- 2026-09-03OpenAI GPT-6 Astra launch chart (vendor, with tools; transcribed by Vellum)
Agentic software engineering: pass@1 on real repo tasks with reported cost per completed task.
Directly comparable
Pricing tiers
Claude Opus 5.5: $4/$20 per MTok, no long-context surcharge. Cache reads $0.20; 5-minute cache writes $5, 1-hour writes $8. Batch $2/$10; Fast mode $8/$40. Batches API allows 300K output with the output-300k-2026-03-24 beta header. Knowledge cutoff Jun 2026; retirement not before 2027-09-22.
GPT-6 Astra: Standard $10/$50 per MTok (≤272K input); above 272K input the whole request bills $20/$75. Cached input $1/MTok; cache writes $12.50 (1.25x input). Batch/Flex $5/$25. Fast mode costs 2x for up to 2x the speed ($20/$100 short context, $40/$150 long). Knowledge cutoff Apr 30 2026. Rolled out from 2026-09-03; Enterprise tenants get it disabled by default.
Claude Opus 5.5
Claude Opus 5.5 is Anthropic's default model for agentic coding and knowledge work — Anthropic says it performs near Fable 5.1 on most work at about 40% lower running cost than Opus 5, and it leads the Artificial Analysis Intelligence Index v4.3.2.
Best for
- Complex agentic coding
- Enterprise knowledge work
- Default frontier model
Watch out
Breaking changes from Opus 5: adaptive thinking cannot be disabled (effort runs low to max, default medium) and forced tool use returns an error. Verbose at max effort, so cap output on cost-sensitive routes.
GPT-6 Astra
GPT-6 Astra is OpenAI's overall flagship — its most capable model for complex reasoning, coding and computer use, and the first it gates at the Preparedness Framework's 'Critical' cybersecurity threshold.
Best for
- Computer-use agents
- Agentic coding
- Hard reasoning
Watch out
Gated at the 'Critical' cyber threshold and disabled by default for Enterprise; long-context work bills $20/$75 above 272K input tokens. GPT-6 Sol ($2/$10) covers most work at a fifth of the price.
Benchmark
DeepSWE 1.1 in context
Only GPT-6 Astra has a published DeepSWE 1.1 result. It is shown against the wider field, with cost per completed task alongside the score.
Local leader
GPT-6 Astra [xhigh]
74%
Rows shown
28
Best published Pass@1 per model (Datacurve's default view)
Snapshot date
2026-09-26
Mirrored from deepswe.datacurve.ai
Better is toward the top-right (higher pass rate, lower cost). X-axis is reversed to match DeepSWE’s public chart. v1.1 uses average cost / tokens / steps; v1 uses published medians.
| # | Model | Pass@1 | Cost / task | Tokens / task | Steps / task |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra [xhigh] | 74% | $4.43 | 30k | 29 |
| 2 | Gemini 3.8 Flash [high] | 74% | $2.36 | 143k | 166 |
| 3 | Claude Opus 5 [max] | 74% | $11.84 | 118k | 99 |
| 4 | GPT-5.6 Sol [max] | 73% | $8.39 | 60k | 61 |
| 5 | Claude Fable 5 [xhigh] | 70% | $13.41 | 80k | 68 |
| 6 | GPT-5.6 Terra [max] | 70% | $4.95 | 72k | 76 |
| 7 | GLM 5.3 [max] | 69% | $3.99 | 80k | 124 |
| 8 | Kimi K3 [max] | 69% | $4.65 | 82k | 98 |
| 9 | Grok 4.6 [medium] | 68% | $3.45 | 50k | 70 |
| 10 | GPT-5.6 Luna [max] | 67% | $3.03 | 73k | 102 |
| 11 | GPT-5.5 [xhigh] | 67% | $7.23 | 46k | 82 |
| 12 | Gemini 3.7 Flash [medium] | 66% | $2.03 | 94k | 117 |
| 13 | GLM 5.3 Flash [max] | 63% | $0.48 | 73k | 123 |
| 14 | DeepSeek V4-Pro [max] | 63% | $0.24 | 106k | 155 |
| 15 | Claude Opus 4.8 [max] | 59% | $13.22 | 135k | 120 |
| 16 | Qwen3.8-Max [xhigh] | 58% | $3.73 | 95k | 111 |
| 17 | Muse Spark 1.2 [xhigh] | 55% | $3.70 | 99k | 101 |
| 18 | Claude Sonnet 5 [max] | 54% | $26.40 | 214k | 268 |
| 19 | Grok 4.5 [high] | 54% | $2.42 | 36k | 61 |
| 20 | DeepSeek V4-Flash [max] | 53% | $0.10 | 108k | 153 |
| 21 | Muse Spark 1.1 [xhigh] | 53% | $2.36 | 74k | 96 |
| 22 | GPT-5.4 [xhigh] | 52% | $5.65 | 71k | 70 |
| 23 | Gemini 3.6 Flash [high] | 47% | $4.42 | 96k | 117 |
| 24 | GLM 5.2 [max] | 44% | $3.92 | 78k | 129 |
| 25 | Gemini 3.5 Flash [high] | 36% | $3.45 | 76k | 105 |
| 26 | Kimi K2.7 Code | 31% | $2.82 | 59k | 149 |
| 27 | Claude Sonnet 4.6 [high] | 30% | $5.52 | 76k | 134 |
| 28 | Gemini 3.1 Pro [high] | 12% | $2.14 | 28k | 76 |
“Best per model” shows each model’s highest published Pass@1, as Datacurve’s own board does; switch to all effort levels to see every configuration. Small gaps may not be statistically meaningful — confirm on deepswe.datacurve.ai.
When the cheaper one wins
Claude Opus 5.5 is cheaper on output at $20 per million tokens against $50 for GPT-6 Astra — about 2.5×. Use the cheaper tier for classification, extraction, summarisation, and any task where the expensive model’s extra score does not change the accepted output. The expensive one only pays if your hardest task actually fails on the cheap tier. These are standard-tier API rates, excluding batch and cache discounts.
Run the model pickerEvidence confidence: High
How strong and complete the evidence behind this comparison is — not a prediction of which model is better.
- Pricing verified on both sides — Input and output rates are verified for both models.
- 4/5 core specs verified on both sides — Not published for at least one side: parameter count.
- 4 shared named benchmarks with differing scores — Measured on: Artificial Analysis Intelligence Index, Humanity's Last Exam, Terminal-Bench 4.0, Terminal-Bench 4.0.
- Verified within the last 90 days — Newest catalog check was 2 days ago.
- Both models carry source citations — Each side has at least two catalog sources on record.
Source receipts
Each catalog figure was checked against the provider or an independent second source on the date shown.
- Claude Opus 5.5: Anthropic — Claude Opus 5.5 (accessed 2026-09-26)
- Claude Opus 5.5: Anthropic docs — Opus 5.5 overview (1M context, 128K output) (accessed 2026-09-26)
- Claude Opus 5.5: Anthropic pricing (claude-opus-5-5 $4/$20, cache reads $0.20) (accessed 2026-09-26)
- Claude Opus 5.5: Artificial Analysis — Intelligence Index v4.3.2 (Opus 5.5 #1) (accessed 2026-09-26)
- GPT-6 Astra: OpenAI — Introducing GPT-6 Astra (accessed 2026-09-03)
- GPT-6 Astra: OpenAI API pricing (gpt-6-astra $10/$50) (accessed 2026-09-03)
- GPT-6 Astra: CNBC — OpenAI launches GPT-6 Astra (accessed 2026-09-03)
- GPT-6 Astra: OpenAI model docs — gpt-6-astra (128K max output, effort levels) (accessed 2026-09-05)
- GPT-6 Astra: OpenAI API pricing (gpt-6-astra ≤272K $10/$50, >272K $20/$75) (accessed 2026-09-26)
Related comparisons
- Claude Fable 5.1 vs Claude Opus 5.5
- Claude Opus 5.5 vs DeepSeek V4.1 Flash
- Claude Opus 5.5 vs Gemini 3.8 Flash
- Claude Opus 5.5 vs GPT-6 Luna
- Claude Opus 5.5 vs GPT-6 Sol
- Claude Opus 5.5 vs Grok 4.7
Diving deeper on one model? Claude Opus 5.5 · GPT-6 Astra
Common questions
Claude Opus 5.5 vs GPT-6 Astra
Answered from the verified figures on this page rather than general guidance.
Is Claude Opus 5.5 or GPT-6 Astra cheaper for input?
Is Claude Opus 5.5 or GPT-6 Astra cheaper for output?
Which has the larger context window, Claude Opus 5.5 or GPT-6 Astra?
Do Claude Opus 5.5 and GPT-6 Astra support the same reasoning levels?
Should I use Claude Opus 5.5 or GPT-6 Astra?
Next step
Choosing between them
The verified figures that separate this pair, computed from the catalog rather than restated boilerplate.
- Input price: Claude Opus 5.5 costs $4 per 1M tokens versus $10 for GPT-6 Astra — a 2.5x difference at the headline tier.
- Context: GPT-6 Astra takes 1.05M against 1M for Claude Opus 5.5 — only decisive if your prompts approach the smaller window.
- Measured capability: Claude Opus 5.5 leads Artificial Analysis Intelligence Index 57.6 to 52.7 (measured 2026-09-26).
Choosing a harness rather than a model — Cursor, Copilot, Claude Code, Windsurf, or Lovable? Compare agentic harnesses · Latest releases.