vrr/qwen-code
2
0
Fork 0
mirror of https://github.com/QwenLM/qwen-code.git synced 2026-06-01 14:21:21 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 10

fix(core): correct GPT-5.x input token limit to 272K

Browse source 
GPT-5.x models have 400K total context but 128K is reserved for output,
so the actual input limit is 272K (400K - 128K). Also updates GLM-5 and
GLM-4.7 output limits from 16K to 128K.

Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com>
This commit is contained in:
tanzhenxin 2026-03-13 14:26:12 +08:00
parent 625c50aae7
commit 613e8b3633
2 changed files with 9 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

10
packages/core/src/core/tokenLimits.test.ts
View file

@ -108,11 +108,11 @@ describe('tokenLimit', () => {
});
describe('OpenAI', () => {
it('should return 400K for GPT-5.x (latest)', () => {
expect(tokenLimit('gpt-5')).toBe(400000);
expect(tokenLimit('gpt-5-mini')).toBe(400000);
expect(tokenLimit('gpt-5.2')).toBe(400000);
expect(tokenLimit('gpt-5.2-pro')).toBe(400000);
it('should return 272K for GPT-5.x (latest)', () => {
expect(tokenLimit('gpt-5')).toBe(272000);
expect(tokenLimit('gpt-5-mini')).toBe(272000);
expect(tokenLimit('gpt-5.2')).toBe(272000);
expect(tokenLimit('gpt-5.2-pro')).toBe(272000);
});
it('should return 128K for legacy GPT (fallback)', () => {

7
packages/core/src/core/tokenLimits.ts
View file

@ -23,6 +23,7 @@ const LIMITS = {
'128k': 131_072,
'200k': 200_000, // vendor-declared decimal, used by OpenAI, Anthropic, etc.
'256k': 262_144,
'272k': 272_000, // vendor-declared decimal, GPT-5.x input (400K total - 128K output)
'400k': 400_000, // vendor-declared decimal, used by OpenAI GPT-5.x
'512k': 524_288,
'1m': 1_000_000,
@ -87,7 +88,7 @@ const PATTERNS: Array<[RegExp, TokenCount]> = [
// -------------------
// OpenAI
// -------------------
[/^gpt-5/, LIMITS['400k']], // GPT-5.x: 400K
[/^gpt-5/, LIMITS['272k']], // GPT-5.x: 272K input (400K total - 128K output)
[/^gpt-/, LIMITS['128k']], // GPT fallback (4o, 4.1, etc.): 128K
[/^o\d/, LIMITS['200k']], // o-series (o3, o4-mini, etc.): 200K
@ -171,8 +172,8 @@ const OUTPUT_PATTERNS: Array<[RegExp, TokenCount]> = [
[/^deepseek-chat/, LIMITS['8k']],
// Zhipu GLM
[/^glm-5/, LIMITS['16k']],
[/^glm-4\.7/, LIMITS['16k']],
[/^glm-5/, LIMITS['128k']],
[/^glm-4\.7/, LIMITS['128k']],
// MiniMax
[/^minimax-m2\.5/i, LIMITS['64k']],