Anthropic Raises Claude API Request Limit to 128 MB | Zeniteq
Loading
Loading article
Zeniteq is now available on mobile for iPhone, iPad and Apple Watch.Download nowLearn more
Anthropic Raises Claude API Request Limit to 128 MB
The fourfold increase gives Messages and Token Counting more room for large payloads, while batches, files, managed agents and model context windows keep separate limits.
Anthropic has increased the maximum request size for its first-party Claude Messages and Token Counting APIs from 32 MB to 128 MB. This update, detailed in an October 28, 2026, release note, allows developers to submit significantly larger payloads to these specific endpoints without hitting previous transport-level ceilings.
The change applies exclusively to the Messages and Token Counting endpoints, while other API routes like Message Batches, Files, and managed-agent sessions retain their existing, separate limits. This adjustment does not expand model context windows, meaning payloads must still satisfy model-specific token requirements. Developers should update application-side validation logic, as client-side gateways or middleware configured for the former 32 MB limit may still reject valid, larger requests.
Anthropic raised the maximum request size for its first-party Claude Messages and Token Counting APIs from 32 MB to 128 MB, according to its October 28, 2026 release note. Developers can now submit substantially larger requests to those two endpoints before hitting the API’s request-size ceiling. (Anthropic release notes)
The change matters most when a request carries large documents, many images or a growing computer-use history. Those payloads can run out of room in megabytes before exhausting the model’s context window. Raising the request limit removes that particular bottleneck for more workloads, but developers still need to check whether the model can process the contents.
The 128 MB limit is endpoint-specific. Message Batches, Files and the APIs for managed-agent Sessions, Agents and Environments retain different ceilings. Treating this as a platform-wide increase would leave some integrations with the wrong validation rules.
What Changed for Messages and Token Counting
The increase covers two first-party endpoints:
Messages:POST /v1/messages, which sends conversational requests to Claude.
Token Counting:POST /v1/messages/count_tokens, which counts a message’s tokens before a developer submits it for inference.
Both move from 32 MB to 128 MB. Anthropic’s announcement explicitly limits the change to the first-party Claude API; developers using cloud-provider-managed endpoints should check that platform’s documentation instead of assuming the same ceiling applies. (Anthropic release notes, API overview)
Raising both limits together is a useful detail. If Messages accepted a larger payload while Token Counting retained the old ceiling, developers could lose their token-checking step precisely when handling bigger inputs. The matched limits allow more large requests to pass through that preflight workflow, although the counting request must still fit within its own 128 MB limit.
The release note describes a request-size increase, without specifying a required SDK upgrade, new API version or beta header for it. That makes the immediate integration task a review of application-side restrictions: a client, gateway or upload form still configured to reject requests above 32 MB could prevent developers from using the extra room.
Anthropic’s API overview lists the following first-party request-size ceilings. Only Messages and Token Counting are identified in the October 28 announcement as receiving the increase. (API overview)
API
Maximum request size
Effect of the announced change
Messages
128 MB
Increased from 32 MB
Token Counting
128 MB
Increased from 32 MB
Message Batches
256 MB
Separate limit, unchanged by this announcement
Files
500 MB
Separate limit, unchanged by this announcement
Sessions
32 MB
Separate limit, unchanged by this announcement
Agents
32 MB
Separate limit, unchanged by this announcement
Environments
32 MB
Separate limit, unchanged by this announcement
These numbers describe different API operations. The Files API uploads and manages files for use across calls. Message Batches processes collections of Messages requests asynchronously. Sessions, Agents and Environments provide different pieces of Anthropic’s managed-agent infrastructure. Their limits should follow the endpoint receiving the request. (API overview)
For example, a hypothetical 40 MB request is below the new Messages ceiling but above the listed Sessions ceiling. An application cannot safely apply a single 128 MB check to both routes.
The 500 MB Files limit also needs careful interpretation. It describes the file-upload endpoint’s capacity; it does not promise that a model can consume every uploaded file in one turn. Similarly, the 32 MB Sessions request ceiling should not be read as a statement about the total amount of state a session can accumulate. The documentation labels these as request-size limits.
I’d keep these values in an endpoint-specific configuration table. The published differences are large enough that one shared “Claude upload limit” would either reject usable requests or allow requests that the destination will refuse.
Request Size and Context Windows Measure Different Things
Request size measures how much data is submitted to an endpoint. A model’s context window governs how much model input it can accommodate, measured in tokens. The October 28 announcement changes the former; it announces no expansion of model context windows. (Anthropic release notes)
The practical consequence is that a request has more than one test to pass:
Does the request fit the endpoint’s size ceiling?
Does the content fit the selected model’s context requirements?
Passing either check says nothing definitive about the other. A hypothetical 60 MB Messages request is below the new 128 MB ceiling, but its contents still need to fit the model’s context and satisfy the API’s other validation rules. Conversely, an image-heavy request could have context capacity available yet previously have exceeded the 32 MB request limit.
Token Counting helps with the second question. Anthropic describes that API as a way to count tokens before sending a message, supporting cost and rate-limit management. It now has the same larger request ceiling as Messages. (API overview)
I’d keep byte-size validation and token validation as distinct checks in an application. They diagnose different problems and suggest different remedies. Removing an oversized attachment might solve a request-size failure; shortening conversation history might address a context-capacity problem. An error message that simply says “input too large” hides that distinction from both users and developers.
Where the Extra Room Helps
Anthropic’s ClaudeDevs account identified large PDFs and long computer-use sessions as examples of workloads affected by the old ceiling. In a post reproduced by Dev Radar, the account said image-heavy requests had been reaching 32 MB while still having context capacity available. That is the vendor’s explanation of the update’s motivation, not an independent workload measurement. (Dev Radar’s reproduction of the ClaudeDevs post)
Three practical benefits follow from the documented increase:
Larger document submissions: Requests previously rejected solely for landing between 32 MB and 128 MB now have room to pass the size check.
More image-heavy history: Computer-use workflows can carry more request data before reaching the transport ceiling, provided the contents remain valid for the model.
Larger token-counting preflights: Developers can assess bigger messages without immediately encountering the former 32 MB counting limit.
The strongest benefit is avoiding premature restructuring. If a useful request was too large in bytes but otherwise fit the model, developers may no longer need to remove images or divide the task solely to satisfy the old ceiling.
That does not make unbounded history a good default. Larger payloads still require transfer and processing, and the announcement offers no performance measurements showing how latency changes near the new maximum. Selective history retention remains sensible when older material no longer helps the task.
Handling a 413 Request Too Large Error
Requests exceeding an endpoint’s ceiling receive HTTP 413, with the error type request_too_large. Anthropic’s documentation says Cloudflare returns this error on the first-party API before the request reaches the API servers. (API overview, recorded documentation change)
That makes a size-related 413 a poor candidate for an unchanged retry. Sending the same oversized payload again does not address the reason it was rejected.
A useful handling sequence is to identify the destination endpoint, check the submitted request’s size, and reduce or restructure it against that endpoint’s ceiling. For Messages and Token Counting, use 128 MB as the published limit. Preserve the different values for other routes.
If splitting the input is necessary, preserve relationships between the pieces. Dividing a document across independent requests can lose cross-document context unless the application deliberately carries forward the information needed for the next step.
For an integration review, the most useful checks are straightforward:
Replace stale 32 MB validation for Messages and Token Counting.
Check application gateways and request-body middleware for smaller local limits.
Keep token-counting and context checks in place.
Record the endpoint and size-related error without logging sensitive payload contents.
The extra capacity is most valuable when it removes an accidental transport constraint while leaving those controls intact.
Frequently Asked Questions
4 questions
1
Which Claude API endpoints now accept 128 MB requests?
The first-party Messages and Token Counting APIs. Anthropic announced the increase from 32 MB on October 28, 2026.
2
Does the increase expand Claude’s context window?
No. Request size and context capacity are separate constraints. A request below 128 MB must still fit the selected model’s context requirements.
3
Do managed-agent endpoints also get the higher limit?
The API overview lists Sessions, Agents and Environments at 32 MB. The announcement covers Messages and Token Counting.
4
What happens if a request exceeds its endpoint’s limit?
The first-party API returns HTTP 413 with request_too_large. Reduce or restructure the payload before retrying it.
