Compare model

Chat

AnthropicClaude Opus 5.5

Deep coding and sustained agent workflows for demanding professional tasks.

claude-opus-5-5

Best for: Complex reasoning, multi-step analysis and difficult coding

ShareLinkedInX

Catalog status

Enabled

Research checked

2026-10-07

Price / performance

What it costs to get this result

FrontierCode v1.1 (Main) · Source-reported results, not a composite ranking. Prices use the average of TokenRex input and output rates per million tokens; actual spend depends on usage. Test configurations differ.

Claude Opus 5.554.4% · 43 sourced models

Blended price / 1M tokens · log scale

● Current model● Other modelsDashed line: price/score frontier among plotted models
Compare sourced models on this benchmark

Benchmark results

17 reported results

Results are source-attributed, not TokenRex measurements. Different tests and thinking variants are not interchangeable.

17 of 17 results

FrontierCode v1.1 (Main)

Coding · Vendor-reported

Test conditions and source

Provider launch evaluation with adaptive thinking at max effort and production safeguards; not the default medium catalog configuration.

Tested model: claude-opus-5-5

Anthropic · Claude Opus 5.5 launch evaluation · Checked 2026-10-06

54.4%

CursorBench 4.0

Coding · Vendor-reported

Test conditions and source

Provider launch evaluation with adaptive thinking at max effort and production safeguards; not the default medium catalog configuration.

Tested model: claude-opus-5-5

Anthropic · Claude Opus 5.5 launch evaluation · Checked 2026-10-06

57.8%

AutomationBench

Agentic · Vendor-reported

Test conditions and source

Provider launch evaluation with adaptive thinking at max effort and production safeguards; not the default medium catalog configuration.

Tested model: claude-opus-5-5

Anthropic · Claude Opus 5.5 launch evaluation · Checked 2026-10-06

40%

Humanity’s Last Exam (with tools)

Reasoning · Vendor-reported

Test conditions and source

Provider launch evaluation with adaptive thinking at max effort and production safeguards; not the default medium catalog configuration.

Tested model: claude-opus-5-5

Anthropic · Claude Opus 5.5 launch evaluation · Checked 2026-10-06

67.7%

Terminal-Bench-Science 0.1

Agentic · Vendor-reported

Test conditions and source

Provider launch evaluation with adaptive thinking at max effort and production safeguards; not the default medium catalog configuration.

Tested model: claude-opus-5-5

Anthropic · Claude Opus 5.5 launch evaluation · Checked 2026-10-06

58.7%

OSWorld 2.1 (partial)

Agentic · Vendor-reported

Test conditions and source

Provider launch evaluation with adaptive thinking at max effort and production safeguards; not the default medium catalog configuration.

Tested model: claude-opus-5-5

Anthropic · Claude Opus 5.5 launch evaluation · Checked 2026-10-06

81.8%

Chartography (with tools)

Reasoning · Vendor-reported

Test conditions and source

Provider launch evaluation with adaptive thinking at max effort and production safeguards; not the default medium catalog configuration.

Tested model: claude-opus-5-5

Anthropic · Claude Opus 5.5 launch evaluation · Checked 2026-10-06

89%

Terminal-Bench 4.0

Agentic · Vendor-reported

Test conditions and source

Launch report explicitly states xhigh effort, standard error ±2.6 points; production safeguards. Not a medium or fast-mode measurement.

Tested model: claude-opus-5-5

Anthropic · Claude Opus 5.5 launch evaluation · Checked 2026-10-06

66.4%

Artificial Analysis Intelligence Index

Reasoning · Third-party report

Test conditions and source

Independent AA run, 'Claude Opus 5.5 (Max, Default Fallback)' configuration — AA's max-effort-with-fallback harness, not the catalog medium default. Not comparable to the Xhigh config (56) or High config (54) also reported by AA.

Tested model: claude-opus-5-5

Artificial Analysis · Claude Opus 5.5 release page · Checked 2026-10-06

58index

Humanity's Last Exam

Reasoning · Third-party report

Test conditions and source

Independent AA evaluation, reported for the top-scoring (max-effort/fallback) AA configuration used for the Intelligence Index; AA states this is the new best score on this eval, ahead of Fable 5.1's prior 59.1%. Effort level for this specific sub-score is implied (Intelligence Index default config), not separately confirmed for every effort tier.

Tested model: claude-opus-5-5

Artificial Analysis · 'Claude Opus 5.5 takes the top spot' article · Checked 2026-10-06

61.4%

SciCode

Coding · Third-party report

Test conditions and source

Independent AA evaluation under the Intelligence Index max-effort/fallback configuration; AA reports this surpasses Fable 5.1's prior best of 63.1%.

Tested model: claude-opus-5-5

Artificial Analysis · 'Claude Opus 5.5 takes the top spot' article · Checked 2026-10-06

66.9%

AA-Briefcase v1.1 (Elo)

Agentic · Third-party report

Test conditions and source

AA's private frontier knowledge-work evaluation using the open-source Stirrup reference agent harness; max-effort/fallback configuration (the configuration AA headlines for Opus 5.5). Not the Xhigh-specific Briefcase score (1768) also published by AA.

Tested model: claude-opus-5-5

Artificial Analysis · 'Claude Opus 5.5 takes the top spot' article · Checked 2026-10-06

1822Elo

Terminal-Bench 4.0

Agentic · Third-party report

Test conditions and source

Explicitly the Xhigh effort configuration ('Claude Opus 5.5 (Xhigh, Default Fallback)') on AA's independently-run Terminal-Bench 4.0 harness — matches the requested Opus 5.5 xhigh condition exactly, not max/fast/default. Distinct from Anthropic's own self-reported xhigh figure of 66.4% already on file (different methodology/harness) and from AA's narrative max-effort figure of 59.6% quoted in AA's launch article.

Tested model: claude-opus-5-5

Artificial Analysis · Opus 5.5 (Xhigh) vs Opus 5 (High) comparison · Checked 2026-10-06

60%

Terminal-Bench-Science 0.1

Agentic · Third-party report

Test conditions and source

Explicitly the xhigh effort configuration ('Claude Opus 5.5 (xhigh)'); AA also separately reports 59% at max effort for the same model on this benchmark — kept distinct and not merged. Independent AA harness, not Anthropic's own 58.7% (max effort) figure already on file.

Tested model: claude-opus-5-5

Artificial Analysis · Terminal-Bench-Science 0.1 leaderboard launch post (LinkedIn) · Checked 2026-10-06

62%

CursorBench 4.0

Coding · Third-party report

Test conditions and source

Row explicitly labeled 'Opus 5.5 Extra High' (xhigh) at 56.0%, $6.98/task. Kept distinct from the existing 57.8% entry, which is Anthropic's own self-reported max-effort figure from the launch page, and from CursorBench's own 'Opus 5.5 Max' row (57.8%, $13.43/task) — same number as Anthropic's but a separately sourced/priced data point.

Tested model: claude-opus-5-5

Cursor · CursorBench 4.0 public leaderboard · Checked 2026-10-06

56%

FrontierCode v1.1 (Main)

Coding · Third-party report

Test conditions and source

Cognition FrontierCode 1.1; Claude Opus 5.5; medium effort; claude-code harness; Main subset (100 tasks). Published weighted rubric score (new_score × 100), not the all-or-nothing pass rate; solution-source violations score zero. Source field new_score=0.5464. Publisher evaluation, not a TokenRex measurement; provider effort configuration is not a guarantee of the API's fixed thinking budget.

Tested model: claude-opus-5-5

Cognition · FrontierCode 1.1 original leaderboard · Checked 2026-10-07

54.64%

FrontierCode v1.1 (Extended)

Coding · Third-party report

Test conditions and source

Cognition FrontierCode 1.1; Claude Opus 5.5; medium effort; claude-code harness; Extended subset (150 tasks). Published weighted rubric score (new_score × 100), not the all-or-nothing pass rate; solution-source violations score zero. Source field new_score=0.6527. Publisher evaluation, not a TokenRex measurement; provider effort configuration is not a guarantee of the API's fixed thinking budget.

Tested model: claude-opus-5-5

Cognition · FrontierCode 1.1 original leaderboard · Checked 2026-10-07

65.27%

Model specifications

Context window

1M

Published provider limit

Maximum output

128K

Published provider limit

Knowledge cutoff

Jun 2026

Upstream model

Input and output modalities

Provider-reported formats; API compatibility is detailed below.

Input

Text
Image

Output

Text
Exact model ID
claude-opus-5-5
Model type
Chat
Context window
1,000,000 tokens
Maximum output
128,000 tokens
Knowledge cutoff
Jun 2026

A context window is the total conversation budget, not a separate maximum input allowance; generated output and reasoning can use that budget. Published provider limits are not independently tested TokenRex request limits.

Versions and thinking levels

Choose from 8 enabled versions. The exact ID determines the version and its price; benchmark scores do not carry over between variants.

Claude Opus 5.5Selected$4.00 in / $20.00 out per 1M tokens
Claude Opus 5.5 Fast$8.00 in / $40.00 out per 1M tokens
Claude Opus 5.5 Fast High$8.00 in / $40.00 out per 1M tokens
Claude Opus 5.5 Fast Low$8.00 in / $40.00 out per 1M tokens
Claude Opus 5.5 Fast Max$8.00 in / $40.00 out per 1M tokens
Claude Opus 5.5 High$4.00 in / $20.00 out per 1M tokens
Claude Opus 5.5 Low$4.00 in / $20.00 out per 1M tokens
Claude Opus 5.5 Max$4.00 in / $20.00 out per 1M tokens

Features and tool support

TokenRex’s exposed interface, not every feature advertised by the provider. “Available” describes the implemented interface, not a successful test of every input or tool.

Reasoning configuration

Adaptive thinking; reasoning and answer share the output budget.

Selected effort: medium (default catalog version). Reasoning level is not a measured intelligence or speed score.

Coding and agent workflows

Text/code generation and customer-managed tool loops.

Coding benchmark results do not establish a hosted terminal, sandbox, autonomous browser or guaranteed task success.

Streaming answers

Available

Incremental text through TokenRex’s chat endpoint; Claude also has native Messages access.

Customer-defined function tools

Limited

Function schemas and tool results are forwarded to Claude. The chat bridge uses automatic tool selection; your application authorizes and executes every tool.

Forced or named tool selection

Not exposed

The chat bridge selects tools automatically. Native Messages does not expose forced or named tool selection; do not assume a tool will be called every turn.

Images and PDFs

Available

Images and PDFs can be sent in chat messages to every chat model. File size and model limits apply.

Hosted web search / browsing

Not exposed

No provider-hosted web search or web-fetch tool is exposed. Customer-owned search can be implemented as an authorized function tool.

Hosted code execution / computer use

Not exposed

No built-in execution environment or computer-control tool is provided. Code generation is not code execution.

Hosted file search / persistent assistants

Not exposed

No hosted file-search index, Assistants endpoint or persistent provider agent is exposed.

Prompt caching

Available

Gateway-reported cache reads and writes are included in billing. See this model’s cached input and cache-write rates; a cache hit is not guaranteed. Output is billed at the standard output rate.

Batch / fine-tuning

Not exposed

No public batch or fine-tuning endpoint is offered here.

Sources and verification

Provider specifications describe the upstream model, not a guarantee of every feature through TokenRex. Prices come from the enabled catalog; benchmark results belong to the exact tested model.

Research notes (1)
  • Published provider limits; maximum-capacity requests have not been independently exercised through TokenRex. Thinking-level variants use the same provider model; reasoning and answer tokens share the output budget.