@blockrun/franklin 3.35.2 → 3.35.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -64,11 +64,21 @@ export declare function clearGatewayModelsCache(): void;
64
64
  * Deliberately ignores the TTL: a stale gateway record still beats no record
65
65
  * at all, and the only callers are fallbacks for models we have no static
66
66
  * entry for. Callers must treat the static tables as authoritative — the
67
- * gateway's own metadata is not reliable (verified 2026-07-20: it reports
68
- * max_output 8192 for claude-haiku-4.5, which Anthropic documents as 64000,
69
- * and 64000 for claude-sonnet-4.6, documented as 128000). It is also not
70
- * enforced: Franklin sends max_tokens 16384 to haiku today and the gateway
71
- * accepts it. Treat this as "better than a blind default", nothing more.
67
+ * gateway's own metadata has been wrong for models we already know. As of
68
+ * 2026-07-20 it reported max_output 8192 for claude-haiku-4.5 (Anthropic
69
+ * documents 64000) and 64000 for claude-sonnet-4.6 (documented 128000).
70
+ * Those two specific numbers are being corrected upstream, so do not treat
71
+ * them as current — the point that outlives them is that the catalog can be
72
+ * wrong, and a wrong value here is not cosmetic.
73
+ *
74
+ * Correction to an earlier note here: these values are NOT inert metadata.
75
+ * The gateway clamps with them — Math.min(request.max_tokens, model.maxOutput)
76
+ * in both the messages and chat/completions handlers — and derives its price
77
+ * quote from the clamped ceiling. An over-cap request is accepted rather than
78
+ * rejected: the handler logs "capping to <limit>" server-side and continues,
79
+ * so from here the clamp is invisible until a reply is long enough to hit it.
80
+ * That is why a short smoke test against a wrongly-capped model looks healthy.
81
+ * Treat this as "better than a blind default", nothing more.
72
82
  */
73
83
  export declare function peekGatewayModel(id: string): GatewayModel | null;
74
84
  /** Test helper — seed the cache without a network call. */
@@ -32,11 +32,21 @@ export function clearGatewayModelsCache() {
32
32
  * Deliberately ignores the TTL: a stale gateway record still beats no record
33
33
  * at all, and the only callers are fallbacks for models we have no static
34
34
  * entry for. Callers must treat the static tables as authoritative — the
35
- * gateway's own metadata is not reliable (verified 2026-07-20: it reports
36
- * max_output 8192 for claude-haiku-4.5, which Anthropic documents as 64000,
37
- * and 64000 for claude-sonnet-4.6, documented as 128000). It is also not
38
- * enforced: Franklin sends max_tokens 16384 to haiku today and the gateway
39
- * accepts it. Treat this as "better than a blind default", nothing more.
35
+ * gateway's own metadata has been wrong for models we already know. As of
36
+ * 2026-07-20 it reported max_output 8192 for claude-haiku-4.5 (Anthropic
37
+ * documents 64000) and 64000 for claude-sonnet-4.6 (documented 128000).
38
+ * Those two specific numbers are being corrected upstream, so do not treat
39
+ * them as current — the point that outlives them is that the catalog can be
40
+ * wrong, and a wrong value here is not cosmetic.
41
+ *
42
+ * Correction to an earlier note here: these values are NOT inert metadata.
43
+ * The gateway clamps with them — Math.min(request.max_tokens, model.maxOutput)
44
+ * in both the messages and chat/completions handlers — and derives its price
45
+ * quote from the clamped ceiling. An over-cap request is accepted rather than
46
+ * rejected: the handler logs "capping to <limit>" server-side and continues,
47
+ * so from here the clamp is invisible until a reply is long enough to hit it.
48
+ * That is why a short smoke test against a wrongly-capped model looks healthy.
49
+ * Treat this as "better than a blind default", nothing more.
40
50
  */
41
51
  export function peekGatewayModel(id) {
42
52
  if (!cache)
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@blockrun/franklin",
3
- "version": "3.35.2",
3
+ "version": "3.35.3",
4
4
  "description": "Franklin Agent — The AI agent with a wallet. Spends USDC autonomously to get real work done. Pay per action, no subscriptions.",
5
5
  "type": "module",
6
6
  "exports": {