Anthropic Launches Claude Opus 5 with Fast Mode Research Preview

Claude Opus 5 is Anthropic's most capable Opus model to date with a standard 1 million token context window and optional fast mode.

Anthropic launched Claude Opus 5 around July 24, 2026. The model establishes a 1 million token context window as both the default and the maximum. There is no smaller context variant available for this model. This configuration supports comprehensive input processing in one go. The model also sets the maximum output at 128 thousand tokens with thinking enabled by default. The design choice ensures that all users receive the full context capacity without options for reduced sizes. Such a configuration supports applications that require maintaining coherence over very long sequences of information. The absence of smaller variants simplifies the user experience by providing maximum capability from the start.

What background led to the Claude Opus 5 introduction?

The new model builds on Claude Opus 4.8 with notable advancements. Anthropic highlights the largest gains in deep reasoning. Additional benefits appear in agentic and long-horizon tasks. Test-time compute scaling also sees enhancements. These areas represent key focuses for the update. Previous models provided the foundation for these developments. The company has iterated on capabilities to address complex user needs. The release reflects ongoing efforts to push boundaries in model performance. The emphasis on agentic tasks aligns with growing demand for models that handle extended sequences of actions and decisions.

Developers have used earlier Opus versions in various enterprise settings. The progression to Claude Opus 5 incorporates feedback on reasoning depth. Long-horizon tasks often involve multiple steps that benefit from improved scaling at test time. Anthropic positions the update as addressing these specific performance areas directly.

What new elements appear in the Claude Opus 5 announcement?

The announcement details the context window size explicitly. Fast mode comes as a research preview available only on the Claude API. Standard speed versions extend to other platforms. Pricing differs between the modes to reflect the speed difference. The fast mode maintains the same model intelligence and capabilities. It focuses solely on increasing the rate of output token generation. Access requires contacting an account manager for the preview. The research preview designation indicates that the feature undergoes further evaluation before broader rollout.

Standard mode receives support across multiple services. This approach balances innovation with controlled access for the accelerated option. The announcement specifies that fast mode applies to Claude Opus 5 without changes to core model behavior.

What technical specifications cover context, output, and pricing?

The context window reaches 1 million tokens without reduction options. Output tokens cap at 128 thousand. Standard pricing stands at five dollars per million input tokens and twenty five dollars per million output tokens. Fast mode sets input at ten dollars and output at fifty dollars per million tokens. The speed multiplier for fast mode reaches up to two point five times the standard output rate. This applies to supported Opus models including the new release. The pricing structure doubles costs for the accelerated variant while preserving all other model attributes.

Pricing and performance details for Claude Opus 5 variants

Model Variant	Input Price (USD per MTok)	Output Price (USD per MTok)	Max Output Tokens	Speed Multiplier
Standard	5	25	128k	1x
Fast	10	50	128k	2.5x

These specifications appear in official documentation from Anthropic. The table above summarizes the key differences between the two modes. Users can select based on whether speed or cost takes priority in their workflows.

How does fast mode deliver its performance benefits?

Fast mode operates as a research preview feature. It targets higher output tokens per second. The change does not impact the quality of responses or the model's core abilities. Users pay a premium for the accelerated generation. The option supports Claude Opus 5 specifically on the designated platform. Contact with an account manager initiates the access process for interested parties. This mechanism allows controlled testing of the speed enhancement in real applications.

Claude Opus 5 is a step-change improvement over Claude Opus 4.8, with the largest gains in deep reasoning, agentic and long-horizon tasks, and test-time compute scaling.Anthropic

What platforms provide access to the new model?

The standard version reaches users through the Claude API. It also appears on Amazon Bedrock. Additional support comes from Google Cloud and Microsoft Foundry. Fast mode stays restricted to the Claude API. This setup allows widespread adoption for standard use cases. The limitation on fast mode keeps the accelerated option under controlled preview conditions. Multiple platforms enable integration into diverse enterprise environments without requiring changes to existing infrastructure.

The distribution strategy supports both individual developers and large organizations. Standard access facilitates immediate experimentation while the preview feature undergoes refinement based on usage data.

1. Claude API for standard and fast modes
2. Amazon Bedrock for standard mode
3. Google Cloud for standard mode
4. Microsoft Foundry for standard mode

What implications emerge for users and the market?

Users in enterprise settings may weigh the higher costs of fast mode against time savings. The improvements in agentic tasks could support more effective automation. Developers might leverage the large context for projects involving extensive data. The dual pricing model creates choices based on application requirements. Broader availability of the standard version facilitates integration across different cloud environments. The research preview status indicates potential for future adjustments. Organizations focused on long running processes stand to gain from the enhanced reasoning capabilities described in the release materials.

Market participants will assess return on investment for the premium tier. The availability on Amazon Bedrock in particular opens pathways for AWS customers to adopt the model within their current setups. This multi platform presence increases overall accessibility.

What steps might follow in model development?

Further iterations could expand on the reasoning gains. The fast mode preview may lead to wider availability after evaluation. Focus areas such as long-horizon tasks will likely continue to receive attention. The release sets expectations for continued progress in frontier models from Anthropic. Stakeholders will monitor updates to pricing and feature access. Additional platforms might receive fast mode support in subsequent phases based on the outcomes of the current research preview.

The emphasis on test time compute scaling suggests ongoing investment in techniques that improve performance during inference. This direction aligns with broader industry trends toward more efficient and capable systems.