none reviewed yet · 2 benchmark observations · latest capture Sep 8, 2026
Inspect the evidenceAnthropic · MODEL PROFILE
Claude Sonnet 4.6
Latency-sensitive interactive UI work where instant first tokens matter more than peak reasoning
Saved editorial fit for your current task lens; not a benchmark result
How scoring worksVerified rates (Sep 8, 2026) · 100 runs / month at your token settings
Adjust workloadAt a glance
9 facts- Provider
- Anthropic
- Type
- API model
- Access
- Anthropic API, AWS Bedrock, Claude Code
- Context
- 1,000,000 tokens
- Maximum output
- 64K tokens
- License / weights
- Proprietary
- Input modalities
- text, image, file
- Output modalities
- text
- Pricing notes
- $3 in / $15 out per 1M - more expensive than the newer Sonnet 5 at $2/$10; cache read $0.30, cache write $3.75
The practical trade-offs
Documented strengths
- The most heavily voted Arena row here (19,843 votes, +/-5) - the tightest confidence interval in the set
- SWE-Rebench 60.7% under a standardised ReAct scaffold, one of very few contamination-resistant results available
- TTFT 1.51 s in non-reasoning mode - near-instant responses
- Design Arena 1300 beats several newer models
Limitations to check
- Costs more than the newer, better Sonnet 5
- 64K max output is half the current Claude standard
- AA Intelligence Index 29 in non-reasoning mode is low
- Legacy status on Anthropic's pricing page
Follow the source
7 source linksEvidence & conditions
7Source links
NoneReviewed records
2Benchmark runs
Sep 8, 2026Latest capture
- Evidence source 1
docs.aws.amazon.comDate not recorded - Evidence source 2
anthropic.comDate not recorded - Evidence source 3
arena.aiDate not recorded - Evidence source 4
benchlm.aiDate not recorded - Evidence source 5
benchlm.aiDate not recorded - Evidence source 6
benchlm.aiDate not recorded - Evidence source 7
artificialanalysis.aiDate not recorded
Benchmark observations
SWE-bench Verified Inspect conditions
- name
- SWE-bench Verified
- value
- 79.6
- unit
- %
- source Url
- https://anthropic.com/claude-sonnet-4-6-system-card
- checked At
- 2026-09-08T19:30:56Z
- attribution
- provider-reported
- configuration
- Section 2.2: 500 tasks, adaptive thinking, max effort; prose says 10 trials, table footnote says 25. This is not the 80.2 prompt-modified result.
Terminal-Bench 2.0 Inspect conditions
- name
- Terminal-Bench 2.0
- value
- 59.1
- unit
- %
- source Url
- https://anthropic.com/claude-sonnet-4-6-system-card
- checked At
- 2026-09-08T19:30:56Z
- attribution
- provider-reported
- configuration
- Section 2.3: Harbor/Terminus-2, no thinking, default max effort; 89 tasks, 5 trials, 1x task resource/time limits.
Source reports, measured results, and editorial assessments are separate evidence types. Source dates do not imply a new benchmark run. Capture dates are shown when recorded; otherwise we say the capture date was not recorded. Recorded dates are compared with this catalogue snapshot (Sep 11, 2026), not today’s date.
Method AIS-V2
How this page is assessed
Quality claims follow Method AIS-V2.
Money never buys rank — commercial relationships cannot change rank, eligibility, evidence quality, or publication timing.
Submit or claim a listing is free and does not skip the editorial queue.
Private · this device only