feat: support claude-opus-4-7 - #4293
Merged
Merged
Conversation
Contributor
WalkthroughThis PR adds support for Claude Opus 4.7 model family across the system by registering new model identifiers, extending cloud provider mappings (AWS and Vertex), updating configuration ratios for caching and model pricing, adding the Changes
Sequence DiagramsequenceDiagram
participant Client
participant Handler as Claude Handler
participant RelayAdapter as Relay Adapter
participant CloudProvider as Cloud Provider (AWS/Vertex)
Client->>Handler: Send request (model=claude-opus-4-7-thinking)
Handler->>Handler: Detect model name & thinking suffix
Handler->>Handler: Trim model to base (claude-opus-4-7)
alt Is Opus 4.7 with Thinking
Handler->>Handler: Set Thinking.Type="adaptive"
Handler->>Handler: Set Thinking.Display="summarized"
Handler->>Handler: Set OutputConfig={"effort":"high"}
Handler->>Handler: Clear Temperature, TopP, TopK
else Other Opus Models
Handler->>Handler: Set Thinking.Type="enabled"
Handler->>Handler: Set BudgetTokens per constraints
Handler->>Handler: Set Temperature=1.0
end
Handler->>RelayAdapter: Forward modified request
RelayAdapter->>CloudProvider: Map model & send request
CloudProvider-->>RelayAdapter: Return response
RelayAdapter-->>Handler: Return response
Handler-->>Client: Return final response
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~30 minutes Possibly related PRs
Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 4
🧹 Nitpick comments (1)
relay/channel/claude/relay-claude.go (1)
156-203: Extract the Opus thinking normalization into one helper.This block now duplicates
relay/claude_handler.go:44-104, and the two paths have already drifted: Line 171 clearsTopPforclaude-opus-4-6here, while the handler path keeps the caller value. A shared helper would keep OpenAI and Claude relay behavior aligned for future Claude model/suffix additions.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@relay/channel/claude/relay-claude.go` around lines 156 - 203, Summary: Duplicate Opus "thinking" normalization logic should be extracted into a single helper and the 4.6/4.7 drift resolved by matching the handler behavior. Create a helper (e.g. NormalizeOpusThinking(model string, req *dto.ClaudeRequest) (newModel string)) and move the duplicated logic that inspects TrimEffortSuffix, "-thinking" suffix, sets claudeRequest.Thinking, OutputConfig, Temperature, TopP, TopK, MaxTokens and Model into it; replace both call sites (relay/channel/claude/relay-claude.go and relay/claude_handler.go) to call this helper, and ensure the helper preserves the caller's TopP for claude-opus-4-6 (i.e., do not unconditionally clear TopP for 4.6) so behavior matches the handler path. Ensure the helper uses model_setting.GetClaudeSettings() and ShouldPreserveThinkingSuffix where currently referenced and returns the possibly trimmed model to assign back to claudeRequest.Model.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@dto/claude.go`:
- Around line 448-455: The Thinking.Display field is currently a non-pointer
string which makes it impossible to distinguish unset vs explicit empty when
parsing and re-marshaling; change the Display declaration in the Thinking struct
(dto/claude.go) from `Display string` to `Display *string` with the same
`json:"display,omitempty"` tag, and update all call sites that set or read
Thinking.Display (e.g., assignments in request construction or tests) to use
pointer values (e.g., common.GetPointer("summarized") or &val) and nil checks
where appropriate so the DTO preserves the optional scalar semantics.
In `@relay/channel/aws/constants.go`:
- Line 21: The entry for "claude-opus-4-7" in awsModelIDMap and its
corresponding key in awsModelCanCrossRegionMap uses an unversioned Bedrock model
ID ("anthropic.claude-opus-4-7") which will fail; update these maps by either
replacing that value with the official versioned model ID (e.g.,
"anthropic.claude-opus-4-7-v1") after confirming via the AWS Bedrock
ListFoundationModels API, or remove the "claude-opus-4-7" entries from
awsModelIDMap and awsModelCanCrossRegionMap until the published model ID is
confirmed to avoid broken API calls.
In `@relay/channel/claude/relay-claude.go`:
- Around line 163-184: The current branch correctly sets claudeRequest.Thinking
to adaptive and OutputConfig.effort for claude-opus-4-7 when enabling the
Thinking adapter, but later logic that maps reasoning overrides
(reasoning_effort, Reasoning.MaxTokens or the medium/high branches) still writes
thinking.Type="enabled" and budget_tokens, which will cause 400s for Opus 4.7;
update the downstream mapping so when trimmedModel or textRequest.Model matches
the "claude-opus-4-7" prefix you route reasoning overrides into
claudeRequest.OutputConfig (set effort to "medium"/"high" or include the
appropriate effort JSON) and do NOT set Thinking.Type="enabled" or
budget_tokens/Reasoning.MaxTokens on claudeRequest; keep the existing behavior
for non-4.7 models (i.e., preserve the existing branches that set
Thinking.Type="enabled" and budget_tokens for other models).
In `@relay/claude_handler.go`:
- Around line 75-98: The Opus 4.7 normalization must run regardless of whether
request.Thinking is nil: detect baseModel via strings.TrimSuffix(request.Model,
"-thinking") and if strings.HasPrefix(baseModel, "claude-opus-4-7") always set
request.Thinking = &dto.Thinking{Type:"adaptive", Display:"summarized"}, set
request.OutputConfig = json.RawMessage(`{"effort":"high"}`) and clear
request.Temperature/request.TopP/request.TopK; otherwise preserve any
client-supplied request.Thinking but keep the existing fallback logic that when
request.Thinking == nil sets request.MaxTokens (use common.GetPointer), computes
BudgetTokens using
model_setting.GetClaudeSettings().ThinkingAdapterBudgetTokensPercentage and
assigns request.Thinking accordingly, and sets default Temperature to
common.GetPointer[float64](1.0).
---
Nitpick comments:
In `@relay/channel/claude/relay-claude.go`:
- Around line 156-203: Summary: Duplicate Opus "thinking" normalization logic
should be extracted into a single helper and the 4.6/4.7 drift resolved by
matching the handler behavior. Create a helper (e.g. NormalizeOpusThinking(model
string, req *dto.ClaudeRequest) (newModel string)) and move the duplicated logic
that inspects TrimEffortSuffix, "-thinking" suffix, sets claudeRequest.Thinking,
OutputConfig, Temperature, TopP, TopK, MaxTokens and Model into it; replace both
call sites (relay/channel/claude/relay-claude.go and relay/claude_handler.go) to
call this helper, and ensure the helper preserves the caller's TopP for
claude-opus-4-6 (i.e., do not unconditionally clear TopP for 4.6) so behavior
matches the handler path. Ensure the helper uses
model_setting.GetClaudeSettings() and ShouldPreserveThinkingSuffix where
currently referenced and returns the possibly trimmed model to assign back to
claudeRequest.Model.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 658bee7a-47c1-4a66-bbc4-c43b14c8927c
📒 Files selected for processing (9)
dto/claude.gorelay/channel/aws/constants.gorelay/channel/claude/constants.gorelay/channel/claude/relay-claude.gorelay/channel/vertex/adaptor.gorelay/claude_handler.gosetting/ratio_setting/cache_ratio.gosetting/ratio_setting/model_ratio.gosetting/reasoning/suffix.go
Comment on lines
448
to
+455
| type Thinking struct { | ||
| Type string `json:"type,omitempty"` | ||
| BudgetTokens *int `json:"budget_tokens,omitempty"` | ||
| // Display controls whether thinking content is returned in the response. | ||
| // Used with adaptive thinking on Claude Opus 4.7+: "summarized" restores | ||
| // the visible summary that was default on Opus 4.6; "omitted" (default on | ||
| // 4.7) suppresses it. Pass-through field from upstream Anthropic API. | ||
| Display string `json:"display,omitempty"` |
Contributor
There was a problem hiding this comment.
Make Thinking.Display a pointer field.
Display is an optional scalar on a request DTO that gets parsed from client JSON and re-marshaled upstream. With string,omitempty, an explicit empty value is indistinguishable from unset, which breaks the DTO contract used elsewhere in this layer.
♻️ Suggested shape
type Thinking struct {
Type string `json:"type,omitempty"`
BudgetTokens *int `json:"budget_tokens,omitempty"`
- Display string `json:"display,omitempty"`
+ Display *string `json:"display,omitempty"`
}request.Thinking.Display = common.GetPointer("summarized")As per coding guidelines, dto/**/*.go: optional scalar fields MUST use pointer types with omitempty, not non-pointer scalars.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@dto/claude.go` around lines 448 - 455, The Thinking.Display field is
currently a non-pointer string which makes it impossible to distinguish unset vs
explicit empty when parsing and re-marshaling; change the Display declaration in
the Thinking struct (dto/claude.go) from `Display string` to `Display *string`
with the same `json:"display,omitempty"` tag, and update all call sites that set
or read Thinking.Display (e.g., assignments in request construction or tests) to
use pointer values (e.g., common.GetPointer("summarized") or &val) and nil
checks where appropriate so the DTO preserves the optional scalar semantics.
| "claude-haiku-4-5-20251001": "anthropic.claude-haiku-4-5-20251001-v1:0", | ||
| "claude-opus-4-5-20251101": "anthropic.claude-opus-4-5-20251101-v1:0", | ||
| "claude-opus-4-6": "anthropic.claude-opus-4-6-v1", | ||
| "claude-opus-4-7": "anthropic.claude-opus-4-7", |
Contributor
There was a problem hiding this comment.
🧩 Analysis chain
🌐 Web query:
site:docs.aws.amazon.com Bedrock "claude-opus-4-7" model ID
💡 Result:
No Bedrock model ID "claude-opus-4-7" exists. Claude Opus 4.7 is listed as available in Amazon Bedrock (https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards-anthropic.html), but its specific model ID is not explicitly stated in model ID lists or model cards. Similar models use IDs like anthropic.claude-opus-4-6-v1 (for Opus 4.6) or anthropic.claude-opus-4-5-20251101-v1:0 (for Opus 4.5). It may follow the pattern anthropic.claude-opus-4-7-v1, but this is not confirmed in official sources. Check the latest AWS Bedrock console or ListFoundationModels API for the exact ID.
Citations:
- 1: https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards-anthropic.html
- 2: https://docs.aws.amazon.com/bedrock/latest/userguide/model-ids.html
- 3: https://docs.aws.amazon.com/bedrock/latest/userguide/models-supported.html
🏁 Script executed:
# First, let's check if the file exists and view the relevant sections
if [ -f relay/channel/aws/constants.go ]; then
echo "=== Lines 15-30 (around line 21) ==="
sed -n '15,30p' relay/channel/aws/constants.go
echo ""
echo "=== Lines 90-105 (around lines 95-99) ==="
sed -n '90,105p' relay/channel/aws/constants.go
echo ""
echo "=== Total line count ==="
wc -l relay/channel/aws/constants.go
else
echo "File not found"
fiRepository: QuantumNous/new-api
Length of output: 1387
Replace model ID with documented versioned format or remove unpublished entry.
The Bedrock model ID anthropic.claude-opus-4-7 breaks the established pattern—all other Opus models in this file use versioned formats (anthropic.claude-opus-4-6-v1, anthropic.claude-opus-4-5-20251101-v1:0). AWS Bedrock documentation lists Opus 4.7 as available but does not explicitly publish the model ID. This unversioned ID will cause API requests to fail. Either use the documented versioned format (likely anthropic.claude-opus-4-7-v1, though unconfirmed) by checking the latest AWS Bedrock ListFoundationModels API, or remove this entry until the official model ID is published.
Affected locations: line 21 (awsModelIDMap) and lines 95–99 (awsModelCanCrossRegionMap).
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@relay/channel/aws/constants.go` at line 21, The entry for "claude-opus-4-7"
in awsModelIDMap and its corresponding key in awsModelCanCrossRegionMap uses an
unversioned Bedrock model ID ("anthropic.claude-opus-4-7") which will fail;
update these maps by either replacing that value with the official versioned
model ID (e.g., "anthropic.claude-opus-4-7-v1") after confirming via the AWS
Bedrock ListFoundationModels API, or remove the "claude-opus-4-7" entries from
awsModelIDMap and awsModelCanCrossRegionMap until the published model ID is
confirmed to avoid broken API calls.
Comment on lines
+163
to
+184
| if strings.HasPrefix(baseModel, "claude-opus-4-7") { | ||
| // Opus 4.7 rejects non-default temperature/top_p/top_k with 400 | ||
| // and defaults display to "omitted"; restore the 4.6 visible summary. | ||
| claudeRequest.Thinking.Display = "summarized" | ||
| claudeRequest.Temperature = nil | ||
| claudeRequest.TopP = nil | ||
| claudeRequest.TopK = nil | ||
| } else { | ||
| claudeRequest.TopP = nil | ||
| claudeRequest.Temperature = common.GetPointer[float64](1.0) | ||
| } | ||
| } else if model_setting.GetClaudeSettings().ThinkingAdapterEnabled && | ||
| strings.HasSuffix(textRequest.Model, "-thinking") { | ||
|
|
||
| // 因为BudgetTokens 必须大于1024 | ||
| if claudeRequest.MaxTokens == nil || *claudeRequest.MaxTokens < 1280 { | ||
| claudeRequest.MaxTokens = common.GetPointer[uint](1280) | ||
| } | ||
| trimmedModel := strings.TrimSuffix(textRequest.Model, "-thinking") | ||
| if strings.HasPrefix(trimmedModel, "claude-opus-4-7") { | ||
| // Opus 4.7 rejects thinking.type="enabled"; use adaptive at high effort. | ||
| claudeRequest.Thinking = &dto.Thinking{Type: "adaptive", Display: "summarized"} | ||
| claudeRequest.OutputConfig = json.RawMessage(`{"effort":"high"}`) | ||
| claudeRequest.Temperature = nil | ||
| claudeRequest.TopP = nil | ||
| claudeRequest.TopK = nil |
Contributor
There was a problem hiding this comment.
Keep Opus 4.7 on adaptive thinking when reasoning overrides are present.
Lines 163-184 switch Opus 4.7 to thinking.type="adaptive" because "enabled" is rejected, but Lines 206-239 still overwrite that with {type:"enabled", budget_tokens:...} for reasoning_effort / reasoning.max_tokens. claude-opus-4-7 and claude-opus-4-7-thinking requests with those fields will still hit the same 400 this branch is trying to avoid. Route those overrides through OutputConfig.effort for 4.7 instead of reusing the older budget-token path.
Possible direction
+isOpus47 := strings.HasPrefix(strings.TrimSuffix(claudeRequest.Model, "-thinking"), "claude-opus-4-7")
+
if textRequest.ReasoningEffort != "" {
switch textRequest.ReasoningEffort {
case "low":
- claudeRequest.Thinking = &dto.Thinking{
- Type: "enabled",
- BudgetTokens: common.GetPointer[int](1280),
- }
+ if isOpus47 {
+ claudeRequest.Thinking = &dto.Thinking{Type: "adaptive", Display: "summarized"}
+ claudeRequest.OutputConfig = json.RawMessage(`{"effort":"low"}`)
+ claudeRequest.Temperature = nil
+ claudeRequest.TopP = nil
+ claudeRequest.TopK = nil
+ } else {
+ claudeRequest.Thinking = &dto.Thinking{
+ Type: "enabled",
+ BudgetTokens: common.GetPointer[int](1280),
+ }
+ }
}
}Apply the same branching to the medium, high, and explicit Reasoning.MaxTokens path.
Also applies to: 206-239
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@relay/channel/claude/relay-claude.go` around lines 163 - 184, The current
branch correctly sets claudeRequest.Thinking to adaptive and OutputConfig.effort
for claude-opus-4-7 when enabling the Thinking adapter, but later logic that
maps reasoning overrides (reasoning_effort, Reasoning.MaxTokens or the
medium/high branches) still writes thinking.Type="enabled" and budget_tokens,
which will cause 400s for Opus 4.7; update the downstream mapping so when
trimmedModel or textRequest.Model matches the "claude-opus-4-7" prefix you route
reasoning overrides into claudeRequest.OutputConfig (set effort to
"medium"/"high" or include the appropriate effort JSON) and do NOT set
Thinking.Type="enabled" or budget_tokens/Reasoning.MaxTokens on claudeRequest;
keep the existing behavior for non-4.7 models (i.e., preserve the existing
branches that set Thinking.Type="enabled" and budget_tokens for other models).
Comment on lines
75
to
98
| if request.Thinking == nil { | ||
| // 因为BudgetTokens 必须大于1024 | ||
| if request.MaxTokens == nil || *request.MaxTokens < 1280 { | ||
| request.MaxTokens = common.GetPointer[uint](1280) | ||
| } | ||
| baseModel := strings.TrimSuffix(request.Model, "-thinking") | ||
| if strings.HasPrefix(baseModel, "claude-opus-4-7") { | ||
| // Opus 4.7 rejects thinking.type="enabled"; use adaptive at high effort. | ||
| request.Thinking = &dto.Thinking{Type: "adaptive", Display: "summarized"} | ||
| request.OutputConfig = json.RawMessage(`{"effort":"high"}`) | ||
| request.Temperature = nil | ||
| request.TopP = nil | ||
| request.TopK = nil | ||
| } else { | ||
| // 因为BudgetTokens 必须大于1024 | ||
| if request.MaxTokens == nil || *request.MaxTokens < 1280 { | ||
| request.MaxTokens = common.GetPointer[uint](1280) | ||
| } | ||
|
|
||
| // BudgetTokens 为 max_tokens 的 80% | ||
| request.Thinking = &dto.Thinking{ | ||
| Type: "enabled", | ||
| BudgetTokens: common.GetPointer[int](int(float64(*request.MaxTokens) * model_setting.GetClaudeSettings().ThinkingAdapterBudgetTokensPercentage)), | ||
| // BudgetTokens 为 max_tokens 的 80% | ||
| request.Thinking = &dto.Thinking{ | ||
| Type: "enabled", | ||
| BudgetTokens: common.GetPointer[int](int(float64(*request.MaxTokens) * model_setting.GetClaudeSettings().ThinkingAdapterBudgetTokensPercentage)), | ||
| } | ||
| // TODO: 临时处理 | ||
| // https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking#important-considerations-when-using-extended-thinking | ||
| request.Temperature = common.GetPointer[float64](1.0) | ||
| } |
Contributor
There was a problem hiding this comment.
Apply the Opus 4.7 normalization even when thinking is already present.
The 4.7-specific rewrite only runs inside if request.Thinking == nil. A request like claude-opus-4-7-thinking with a client-supplied thinking object skips the adaptive/display conversion and leaves temperature/top_p/top_k untouched, so the alias can still hit the upstream 400s this branch is trying to avoid.
🐛 Minimal fix sketch
- if request.Thinking == nil {
- baseModel := strings.TrimSuffix(request.Model, "-thinking")
- if strings.HasPrefix(baseModel, "claude-opus-4-7") {
- request.Thinking = &dto.Thinking{Type: "adaptive", Display: "summarized"}
- request.OutputConfig = json.RawMessage(`{"effort":"high"}`)
- request.Temperature = nil
- request.TopP = nil
- request.TopK = nil
- } else {
+ baseModel := strings.TrimSuffix(request.Model, "-thinking")
+ if strings.HasPrefix(baseModel, "claude-opus-4-7") {
+ if request.Thinking == nil {
+ request.Thinking = &dto.Thinking{}
+ }
+ request.Thinking.Type = "adaptive"
+ request.Thinking.Display = "summarized"
+ request.OutputConfig = json.RawMessage(`{"effort":"high"}`)
+ request.Temperature = nil
+ request.TopP = nil
+ request.TopK = nil
+ } else if request.Thinking == nil {
// 因为BudgetTokens 必须大于1024
if request.MaxTokens == nil || *request.MaxTokens < 1280 {
request.MaxTokens = common.GetPointer[uint](1280)📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| if request.Thinking == nil { | |
| // 因为BudgetTokens 必须大于1024 | |
| if request.MaxTokens == nil || *request.MaxTokens < 1280 { | |
| request.MaxTokens = common.GetPointer[uint](1280) | |
| } | |
| baseModel := strings.TrimSuffix(request.Model, "-thinking") | |
| if strings.HasPrefix(baseModel, "claude-opus-4-7") { | |
| // Opus 4.7 rejects thinking.type="enabled"; use adaptive at high effort. | |
| request.Thinking = &dto.Thinking{Type: "adaptive", Display: "summarized"} | |
| request.OutputConfig = json.RawMessage(`{"effort":"high"}`) | |
| request.Temperature = nil | |
| request.TopP = nil | |
| request.TopK = nil | |
| } else { | |
| // 因为BudgetTokens 必须大于1024 | |
| if request.MaxTokens == nil || *request.MaxTokens < 1280 { | |
| request.MaxTokens = common.GetPointer[uint](1280) | |
| } | |
| // BudgetTokens 为 max_tokens 的 80% | |
| request.Thinking = &dto.Thinking{ | |
| Type: "enabled", | |
| BudgetTokens: common.GetPointer[int](int(float64(*request.MaxTokens) * model_setting.GetClaudeSettings().ThinkingAdapterBudgetTokensPercentage)), | |
| // BudgetTokens 为 max_tokens 的 80% | |
| request.Thinking = &dto.Thinking{ | |
| Type: "enabled", | |
| BudgetTokens: common.GetPointer[int](int(float64(*request.MaxTokens) * model_setting.GetClaudeSettings().ThinkingAdapterBudgetTokensPercentage)), | |
| } | |
| // TODO: 临时处理 | |
| // https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking#important-considerations-when-using-extended-thinking | |
| request.Temperature = common.GetPointer[float64](1.0) | |
| } | |
| baseModel := strings.TrimSuffix(request.Model, "-thinking") | |
| if strings.HasPrefix(baseModel, "claude-opus-4-7") { | |
| if request.Thinking == nil { | |
| request.Thinking = &dto.Thinking{} | |
| } | |
| request.Thinking.Type = "adaptive" | |
| request.Thinking.Display = "summarized" | |
| request.OutputConfig = json.RawMessage(`{"effort":"high"}`) | |
| request.Temperature = nil | |
| request.TopP = nil | |
| request.TopK = nil | |
| } else if request.Thinking == nil { | |
| // 因为BudgetTokens 必须大于1024 | |
| if request.MaxTokens == nil || *request.MaxTokens < 1280 { | |
| request.MaxTokens = common.GetPointer[uint](1280) | |
| } | |
| // BudgetTokens 为 max_tokens 的 80% | |
| request.Thinking = &dto.Thinking{ | |
| Type: "enabled", | |
| BudgetTokens: common.GetPointer[int](int(float64(*request.MaxTokens) * model_setting.GetClaudeSettings().ThinkingAdapterBudgetTokensPercentage)), | |
| } | |
| // TODO: 临时处理 | |
| // https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking#important-considerations-when-using-extended-thinking | |
| request.Temperature = common.GetPointer[float64](1.0) | |
| } |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@relay/claude_handler.go` around lines 75 - 98, The Opus 4.7 normalization
must run regardless of whether request.Thinking is nil: detect baseModel via
strings.TrimSuffix(request.Model, "-thinking") and if
strings.HasPrefix(baseModel, "claude-opus-4-7") always set request.Thinking =
&dto.Thinking{Type:"adaptive", Display:"summarized"}, set request.OutputConfig =
json.RawMessage(`{"effort":"high"}`) and clear
request.Temperature/request.TopP/request.TopK; otherwise preserve any
client-supplied request.Thinking but keep the existing fallback logic that when
request.Thinking == nil sets request.MaxTokens (use common.GetPointer), computes
BudgetTokens using
model_setting.GetClaudeSettings().ThinkingAdapterBudgetTokensPercentage and
assigns request.Thinking accordingly, and sets default Temperature to
common.GetPointer[float64](1.0).
seefs001
approved these changes
Apr 17, 2026
prnake
added a commit
to prnake/new-api
that referenced
this pull request
Apr 17, 2026
This reverts commit 47d7bca.
4 tasks
This was referenced Apr 17, 2026
xyfacai
pushed a commit
to xyfacai/new-api
that referenced
this pull request
Apr 20, 2026
* feat: support claude-opus-4-7 * feat: summarized display for opus 4.7
mygoda
pushed a commit
to mygoda/new-api
that referenced
this pull request
Apr 20, 2026
* feat: support claude-opus-4-7 * feat: summarized display for opus 4.7
liuyaaixxa
pushed a commit
to liuyaaixxa/xnew-api
that referenced
this pull request
Apr 23, 2026
* feat: support claude-opus-4-7 * feat: summarized display for opus 4.7
11 tasks
isboyjc
pushed a commit
to isboyjc/amux-api
that referenced
this pull request
Apr 29, 2026
* feat: support claude-opus-4-7 * feat: summarized display for opus 4.7
Jinxuans
referenced
this pull request
in TokFlux-Org/TokFlux
May 9, 2026
* feat: support claude-opus-4-7 * feat: summarized display for opus 4.7
ljdzxx
pushed a commit
to ljdzxx/new-api
that referenced
this pull request
May 9, 2026
* feat: support claude-opus-4-7 * feat: summarized display for opus 4.7
11 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
📝 变更描述 / Description
support claude-opus-4-7
🚀 变更类型 / Type of change
🔗 关联任务 / Related Issue
✅ 提交前检查项 / Checklist
Bug fix，我已提交或关联对应 Issue，且不会将设计取舍、预期不一致或理解偏差直接归类为 bug。📸 运行证明 / Proof of Work
(请在此粘贴截图、关键日志或测试报告，以证明变更生效)
Summary by CodeRabbit
Release Notes