Public benchmarkExternal source reviewReviewed by DanPublished Source checked Updated
Claude Opus 5.5: public coding benchmark receipts
The report
Verdict
Published benchmark evidence
Terminal-Bench 4.0, CursorBench 4.0 and Cognition FrontierCode 2.0 examine different aspects of coding. Their percentages do not combine into one reliable Opus success rate.
On the tier list: The November 17 tier is Dan’s own ranking; public benchmark numbers are a separate evidence channel, not an automatic tier adjustment.
What the benchmarks report
- Terminal-Bench 4.0: 66.4%. Opus 5.5; xhigh; command-line task benchmark. Reported by Anthropic vendor table.
- FrontierCode 2.0 — Cognition: 47.3%. Opus 5.5; max; Claude Code; 132-task suite. Reported by Cognition benchmark operator.
- CursorBench 4.0: 57.8%. Opus 5.5; multi-file coding. Reported by Anthropic vendor table.
Original sources
Source 1 — Terminal-Bench 4.0, CursorBench 4.0.
Source 2 — FrontierCode 2.0 — Cognition.
What this evidence tells a reader
Terminal-Bench 4.0's 66.4% concerns command-line work under the published xhigh setting. CursorBench 4.0's 57.8% concerns multi-file coding. Cognition's 47.3% FrontierCode result comes from a 132-task code-change suite using Claude Code at max effort. These are different tasks and settings, so the numerical spread does not show that Opus performs worse when a job lasts longer.
Comparing the Cognition result with Fable's 50.2% from the same operator raises a useful follow-up question, but does not settle an Opus-versus-Fable choice: the listed reasoning efforts and potentially other run parameters differ. An engineer who needs safe changes should weigh actual patch acceptance and unwanted edits at least as heavily as one leaderboard number.
DanManREAL has a separate Opus-versus-Terra blind project comparison. It concerns actual design choices in that particular workshop task, not this report's public coding benchmarks. Readers should examine each evidence type for the question it can answer.
How to use these results
- For command-line agents, first inspect the exact Terminal-Bench version and effort, rather than comparing unlike coding suites.
- For repository work, measure accepted patches, review time, and out-of-scope edits in your own harness.
- Use the linked firsthand blind trial for its actual design task only; do not transfer its verdict to every coding benchmark.
Limits of this report
- The 66.4% and 57.8% figures come from Anthropic's model-publisher table, while 47.3% is attributed to Cognition.
- This review does not measure Opus costs, correctness or recovery behavior in an identical local task against Fable.
Follow the lab
New test reports and tier-list changes get announced in my Discord's #news channel. The full videos are on YouTube.
Receipts
DanManREAL · AI lab
Test receipt
Claude Opus 5.5
- Terminal-Bench 4.0
- 66.4%Anthropic vendor table (model publisher). Opus 5.5; xhigh; command-line task benchmark. Not independently rerun by DanManREAL.
- FrontierCode 2.0 — Cognition
- 47.3%Cognition benchmark operator (independent operator or secondary publication). Opus 5.5; max; Claude Code; 132-task suite. Not independently rerun by DanManREAL.
- CursorBench 4.0
- 57.8%Anthropic vendor table (model publisher). Opus 5.5; multi-file coding. Not independently rerun by DanManREAL.
- Source review
- November 17, 2026Date these source claims were reviewed; not a claim about when the published benchmark was executed.
Verdict
Published benchmark evidence
Tier list The November 17 tier is Dan’s own ranking; public benchmark numbers are a separate evidence channel, not an automatic tier adjustment.
danmanreal.com/tests/claude-opus-5-5-public-benchmarks
Models covered by external benchmarks
On the tier list's bench
- Claude Opus 5.5 · 5 receipts on file
Related
- AI Coding Agents Tier List
- GPT-6.1 Terra vs Opus 5.5: three real projects, and the designs I picked blind
Independent testing. Sponsorships are always disclosed, and verdicts are never for sale.
Disagree? Bring your receiptsAll reports
- One inbox for my AI coding agents: triage took 34% less time, resuming work did not improveHead-to-headNov 20
- Mythos 5.2: research and security benchmark source reviewPublic benchmarkNov 17
- Fable 5.2: agentic coding benchmark source reviewPublic benchmarkNov 17
- Claude Opus 5.5: public coding benchmark receiptsPublic benchmarkNov 17
- Astra 6.2: new model and public task-evaluation evidencePublic benchmarkNov 17
- Sol 6.2: new model and public coding benchmark evidencePublic benchmarkNov 17
- Gemini 4 Carbon: benchmark source and usage profilePublic benchmarkNov 17
- Qwen 4: Qwen4-Max launch and benchmark sourcesPublic benchmarkNov 17
- Kimi K3.1 Pro: public coding benchmark receiptsPublic benchmarkNov 17
- GLM-5.3: public coding benchmark receiptsPublic benchmarkNov 17
- Gemini 4 Argon: public coding benchmark receiptsPublic benchmarkNov 17
- Claude Sonnet 5.5: public coding benchmark receiptsPublic benchmarkNov 17
- DeepSeek V4.1 Flash: public coding benchmark receiptsPublic benchmarkNov 17
- Muse Wonder Full 1.0: public coding benchmark receiptsPublic benchmarkNov 17
- Grok 4.7: separate release and public coding resultsPublic benchmarkNov 17
- ai-design-check on two of my FABLE screens: 10 warnings, 4 of them false alarmsField testNov 2
- Vercel Agent Desktop used a sample Trip app in 6 actions without moving the mouseField testNov 1
- Cerberus Guard on sample data: one crawler name, two verdicts, and a block earned by behaviorField testOct 31
- GPT-6.1 Terra vs Opus 5.5: three real projects, and the designs I picked blindHead-to-headOct 29
- Steps → Reduced: opening a terminal agent went from 98 seconds to 11AuditOct 28
- One ChatGPT chat directed Opus and Codex at once through DMR_REMOTEField testOct 28
- Claude Duet ran my channel for a week and caught 269 fake AI crawlersField testOct 27
- Kimi K3.1 Pro built my animation family, after a quota stop and 21 minutes of failed requestsField testOct 24
- ChatGPT + Codex built 33 new DMR_REMOTE toolsField testOct 23
- Gemini 4 Argon reviewed a finished episode: 14 clips, one flag, one inconsistencyField testOct 21
- Qwen 3.8 Flash Next on my RTX 5080: 81 tokens a second with 202K tokens readAuditOct 21
- Does an extra AI review step pay off? A small DeepSeek trial said noHead-to-headOct 19
- My real DeepSeek bill: 12.31 billion tokens for $122.28Cost receiptOct 17
- Gemini 4 Argon kept hanging in Gemini CLI. The cause was a 60-second timeoutIncidentOct 14
- I asked 10 AI coding agents for their biggest weaknessSelf-reportOct 14
- Jarvis 27B on my own PC: 30 tokens a second, and it still recalls at 251KAuditOct 10
- My agents' first game benchmark was wrong. Here's how it was caughtIncidentOct 10
- Pixal3D + Blender turned my 2D art into animated 3D charactersField testSep 22
- DeepSeek Flash answered 200 OK and sent nothing for two hoursIncidentSep 14
- Grok, Codex and DeepSeek played WorldOS for Rank 1. Grok won with 98Head-to-headAug 20
- DeepSeek V4 Flash audited 18 of my real coding sessions. Two said 'done' too earlyAuditAug 4