Skip to content
All systems operational0 AI providers monitored, polled every 2 minutes
Live status

Claude Opus 5.5 vs Claude Opus 5

Claude Opus 5.5 shipped on September 22, 2026, and unlike the Opus 4.8 to Opus 5 step, which held the price, this upgrade also cuts it. Opus 5.5 lists $4 per million input tokens and $20 output against $5 and $25 for Opus 5, 20 percent cheaper on both sides of the meter. Cache reads fall further, from $0.50 to $0.20 per million, a 60 percent cut, because Opus 5.5 bills cache hits at 0.05x base input where other Claude models use 0.1x. Anthropic's headline claim is that Opus 5.5 performs at the level of Claude Fable 5.1 on most work and costs 40 percent less to run than Opus 5, with output more than 30 percent faster. The independent numbers agree on direction. The Vals Index puts Opus 5.5 first at 69.69 percent against 67.21 for Opus 5, Artificial Analysis scores it 58 at max effort, first on its Intelligence Index, and Cognition's FrontierCode v1.1 Main has it at 54.6 against 53.4, a real but narrow lead. On Anthropic's own launch table Terminal-Bench 4.0 moves from 52.3 to 66.4, and Vals measured Opus 5.5 on that benchmark independently at 61.62. The porting cost is behavioral rather than financial. Default effort drops from high to medium, thinking is adaptive and can no longer be disabled, forced tool_choice of any or tool returns a 400, thinking blocks are tied to the model and conversation, and computer use only runs through computer_toolset_20260801. None of that forces a move: Opus 5 is not deprecated, stays at $5/$25, and Anthropic lists its retirement as no sooner than July 24, 2027.

Head-to-Head Specs

SpecClaude Opus 5.5Claude Opus 5
ProviderAnthropicAnthropic
Input Price$4.00/1M$5.00/1M
Output Price$20.00/1M$25.00/1M
Context Window1M1M
Released2026-092026-07
Capabilitiestext, vision, tool-use, code, reasoningtext, vision, tool-use, code, reasoning

Benchmark Scores

BenchmarkClaude Opus 5.5Claude Opus 5Winner
FrontierCode v1.154.653.4Claude
Terminal-Bench 4.061.645.5Claude
Humanity's Last Exam (tools)67.763.6Claude

See the full benchmark leaderboard for all models.

Category Breakdown

Price per 1M tokensClaude Opus 5.5

$4/$20 against $5/$25, 20 percent cheaper on input and output

Cache readsClaude Opus 5.5

$0.20 per 1M against $0.50, a 60 percent cut from the 0.05x multiplier on base input

Aggregate index (Vals Index)Claude Opus 5.5

69.69 percent and first overall against 67.21 for Opus 5, independently measured

Agentic coding (FrontierCode v1.1)Claude Opus 5.5

Cognition measures 54.6 against 53.4, a lead of 1.2 points

Agentic terminal coding (Terminal-Bench 4.0)Claude Opus 5.5

66.4 against 52.3 on Anthropic's launch table; Vals independently measures Opus 5.5 at 61.62

Humanity's Last Exam with toolsClaude Opus 5.5

67.7 against 63.6 for Opus 5 on the same Anthropic table, both vendor-run

Computer use (OSWorld 2.0, re-run)Claude Opus 5.5

Anthropic re-ran both on the September 10, 2026 task files with a new harness, 81.8 against 74.0 on partial credit; neither figure compares with the 70.6 Opus 5 launch score

Thinking controlClaude Opus 5

Opus 5 accepts disabled thinking at effort high or below; on Opus 5.5 thinking is always on, and the default effort drops from high to medium

Drop-in compatibilityClaude Opus 5

Forced tool_choice any or tool returns a 400 on Opus 5.5 and computer use requires computer_toolset_20260801, so working Opus 5 code can fail on a straight model swap

Context windowTieTie

Both ship 1 million tokens with 128K output; Opus 5.5 reaches 300K output on Batch with a beta header

Choose Claude Opus 5.5 when:

  • ▸New Opus workloads, since it is cheaper per token and ahead on every benchmark both models share
  • ▸Agent sessions that replay a large cached prefix, where the 60 percent cache-read cut compounds
  • ▸Long-horizon terminal and coding agents, given the Terminal-Bench 4.0 gap
  • ▸Latency-sensitive routes that gain from faster output and a medium default effort
View Claude Opus 5.5 details

Choose Claude Opus 5 when:

  • ▸Pipelines that disable thinking for latency, which Opus 5.5 no longer allows
  • ▸Code that forces tool_choice any or tool, which returns a 400 on Opus 5.5
  • ▸Computer-use harnesses not yet moved to computer_toolset_20260801
  • ▸Workloads with evals tuned to Opus 5 at high effort that cannot be re-tested yet
View Claude Opus 5 details

Frequently Asked Questions

Which is better, Claude Opus 5.5 or Claude Opus 5?

It depends on your use case. Claude Opus 5.5 from Anthropic excels at new opus workloads, since it is cheaper per token and ahead on every benchmark both models share, while Claude Opus 5 from Anthropic is better for pipelines that disable thinking for latency, which opus 5.5 no longer allows. See the full comparison above for detailed benchmarks and pricing.

How much does Claude Opus 5.5 cost compared to Claude Opus 5?

Claude Opus 5.5 costs $4.00 input and $20.00 output per 1M tokens. Claude Opus 5 costs $5.00 input and $25.00 output per 1M tokens.

What is the context window difference between Claude Opus 5.5 and Claude Opus 5?

Claude Opus 5.5 supports 1M tokens, while Claude Opus 5 supports 1M tokens.

More Comparisons

Interactive Compare ToolAll ModelsFull Pricing Guide