Anthropic may be preparing to release Claude Opus 5.5, a new version of its flagship Opus model, according to reports circulating in the AI developer community.
The unreleased model is reportedly being tested under the internal identifier claude-wafer-eap, with multiple reports pointing to September 22, 2026 as a possible launch date.
What is Claude Wafer EAP?
Reports suggest that claude-wafer-eap is an early-access version of what could eventually launch as Claude Opus 5.5.
The “EAP” suffix appears to indicate an early-access or testing program, suggesting Anthropic may currently be evaluating the model with a limited group of users before a wider release.
Early reports associate the model particularly with improvements in:
- software development and coding
- long-running agentic tasks
- complex reasoning
- frontend and visual development
- 3D and interactive web generation
These capability claims remain unverified until Anthropic publishes official evaluations or independent benchmarks become available.
Rumored Claude Opus 5.5 pricing
One of the most interesting parts of the leak is the reported API pricing.
| Model | Input | Output |
|---|---|---|
| Claude Opus 5 | $5 / 1M tokens | $25 / 1M tokens |
| Claude Opus 5.5 — rumored | $4 / 1M tokens | $20 / 1M tokens |
If accurate, Claude Opus 5.5 would be approximately 20% cheaper per token than Opus 5 despite being a newer model.
Anthropic officially prices Claude Opus 5 at $5 per million input tokens and $25 per million output tokens. The alleged Opus 5.5 prices have not been confirmed by Anthropic.
For coding agents, a price reduction could be significant because autonomous development workflows can consume large quantities of both input and output tokens across repeated tool calls.
Is Claude Opus 5.5 available yet?
Not publicly, as of the time of writing.
Anthropic's official documentation continues to list Claude Opus 5 as the current Opus model.
Claude Opus 5 launched on July 24, 2026 and is currently available through the Claude API and Anthropic's consumer products. Anthropic describes it as a model designed for software engineering, reasoning and professional knowledge work.
There is currently no official claude-opus-5-5 API model identifier.
What about Claude Code?
If Opus 5.5 launches, one of the biggest questions will be how quickly it becomes available inside Claude Code.
Opus models are particularly relevant to coding agents because model quality affects more than individual code completions. An agent repeatedly has to:
- understand a repository,
- decide which files and tools to use,
- make changes,
- execute tests,
- diagnose failures,
- verify its implementation.
Even relatively small improvements in reasoning, tool use and token efficiency can compound across these multi-step workflows.
The rumored lower pricing would therefore be especially interesting if Opus 5.5 becomes available to API and Claude Code users at launch.
What is actually confirmed?
At the moment, the situation can be summarized simply.
On the record
- Claude Opus 5 is Anthropic's current publicly available Opus model.
- Opus 5 costs $5 per million input tokens and $25 per million output tokens.
- Anthropic continues to develop and test new frontier Claude models.
Reported, unconfirmed
- Claude Opus 5.5 is being tested.
- Its early-access identifier is
claude-wafer-eap. - September 22 is the intended release date.
- Pricing will fall to $4/M input and $20/M output.
- It provides substantial improvements for coding and long-running agents.
Until Anthropic publishes an announcement or updates its API documentation, those details should be treated as leaks rather than confirmed specifications.
Why Claude Opus 5.5 matters
The most interesting part of this potential release may not simply be higher benchmark scores.
If Anthropic can simultaneously improve agentic coding performance and reduce the price of Opus, it could make its highest-end models practical for a larger portion of everyday software-development workflows.
That would matter particularly as AI coding shifts from short code generation toward autonomous agents capable of working on repositories for extended periods.
For now, Claude Opus 5.5 remains an unannounced model reportedly undergoing early testing.
Last updated