@oh-my-pi/pi-catalog 18.8.1 → 18.8.3

package/CHANGELOG.md

@@ -2,6 +2,18 @@
 
 ## [Unreleased]
 
+## [18.8.3] - 2026-10-07
+
+### Added
+
+- Added Claude Haiku 5.5 with adaptive thinking (low through max effort), image input, a 1M-token context window, 128K output, and its tiered pricing above 100K input tokens.
+
+## [18.8.2] - 2026-10-07
+
+### Fixed
+
+- Fixed Anthropic requests carrying too many inline screenshot bytes by exposing a provider image-byte budget, applied only on the official endpoint ([#14453](https://github.com/can1357/oh-my-pi/issues/14453)).
+
 ## [18.8.1] - 2026-10-07
 
 ### Fixed

package/dist/types/compat/request-size.d.ts

@@ -0,0 +1,7 @@
+import type { Model } from "../types.js";
+/**
+ * Maximum encoded inline-image bytes retained per request for a deployment with a documented body-size limit.
+ * The axis is a host contract: a rerouted `baseUrl` (proxy, gateway) imposes its own limit, so the budget
+ * applies only while the model targets its official endpoint.
+ */
+export declare function resolveInlineImageByteBudget(model: Model): number | undefined;

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.8.1",
+	"version": "18.8.3",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"keywords": [
 		"ai",
@@ -93,12 +93,12 @@
 		"gen:proto": "bun scripts/generate-protocols.ts"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.8.1",
-		"@oh-my-pi/pi-utils": "18.8.1"
+		"@oh-my-pi/omptype": "18.8.3",
+		"@oh-my-pi/pi-utils": "18.8.3"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@oh-my-pi/pi-ai": "18.8.1",
+		"@oh-my-pi/pi-ai": "18.8.3",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/axes.ts

@@ -395,6 +395,7 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	/** How the model line bills an input image; shape and formulas in `./image-tokenization`. */
 	"image-tokenization": { key: "imageTokenization", set: "catalog", shape: "object" },
 	"image-model": { key: "imageModel", set: "catalog", shape: "scalar" },
+	"inline-image-byte-budget": { key: "inlineImageByteBudget", set: "catalog", shape: "scalar" },
 	"limits-patch": { key: "limitsPatch", set: "catalog", shape: "object" },
 	"long-context-cost": { key: "longContext", set: "catalog", shape: "object" },
 	"prompt-cache": { key: "promptCache", set: "catalog", shape: "object" },

package/src/compat/request-size.ts

@@ -0,0 +1,25 @@
+import { toModelSpec } from "../provider-models/bundled-references";
+import type { Model } from "../types";
+import { resolveCatalogAxes } from "./resolve";
+
+const imageBudgetCache = new Map<string, number | null>();
+const IMAGE_BUDGET_CACHE_MAX = 8192;
+
+/**
+ * Maximum encoded inline-image bytes retained per request for a deployment with a documented body-size limit.
+ * The axis is a host contract: a rerouted `baseUrl` (proxy, gateway) imposes its own limit, so the budget
+ * applies only while the model targets its official endpoint.
+ */
+export function resolveInlineImageByteBudget(model: Model): number | undefined {
+	const compat = model.compat;
+	if (!compat || !("officialEndpoint" in compat) || !compat.officialEndpoint) return undefined;
+	const key = `${model.provider} ${model.id} ${model.api}`;
+	let budget = imageBudgetCache.get(key);
+	if (budget === undefined) {
+		const value = resolveCatalogAxes(toModelSpec(model)).inlineImageByteBudget;
+		budget = typeof value === "number" && Number.isFinite(value) && value > 0 ? value : null;
+		if (imageBudgetCache.size >= IMAGE_BUDGET_CACHE_MAX) imageBudgetCache.clear();
+		imageBudgetCache.set(key, budget);
+	}
+	return budget ?? undefined;
+}

package/src/compat/resolve.ts

@@ -107,11 +107,12 @@ class IdentityFacts {
 		return this.is("kimi") && this.family("k2.7-code", "k3");
 	}
 
-	/** Adaptive-thinking Claude generation floor (Opus ≥ min; Sonnet/Fable/Mythos ≥ 5). */
+	/** Adaptive-thinking Claude generation floor (Opus ≥ min; Sonnet/Fable/Mythos ≥ 5; Haiku ≥ 5.5). */
 	anthropicAdaptiveGenAtLeast(opusMin: string): boolean {
 		if (!this.is("anthropic")) return false;
 		if (this.family("opus")) return this.revGte(opusMin);
 		if (this.family("sonnet", "fable", "mythos")) return this.revGte("5");
+		if (this.family("haiku")) return this.revGte("5.5");
 		return false;
 	}
 }
@@ -1009,7 +1010,8 @@ function defaultThinkingMode<TApi extends Api>(spec: ModelSpec<TApi>, facts: Ide
 				return "anthropic-budget-effort";
 			}
 			if (facts.is("anthropic")) {
-				if (facts.revGte("4.6") && !facts.family("haiku")) return "anthropic-adaptive";
+				// Haiku stays on budget thinking until 5.5, its first adaptive generation.
+				if (facts.revGte(facts.family("haiku") ? "5.5" : "4.6")) return "anthropic-adaptive";
 				if (facts.family("opus") && facts.revGte("4.5")) return "anthropic-budget-effort";
 			}
 			return "budget";

package/src/compat/rules/README.md

@@ -217,6 +217,8 @@ One object axis carries a computed form: `long-context-cost` accepts either the
 
 `context-window-authoritative #true` preserves a host's supplied context window through runtime model selection instead of applying inferred expansion or reference-price-tier caps. Explicit user context overrides still apply afterward. It applies to rows materialized through `buildModel` (discovery and regenerated bundles). See [`providers/factory-droid.kdl`](providers/factory-droid.kdl).
 
+`inline-image-byte-budget` caps retained base64 image characters for a deployment's request body. It leaves room for tool schemas and text beneath the provider's body-size limit; the coding agent reads it through `resolveInlineImageByteBudget` before sending live history. The budget applies only while `compat.officialEndpoint` holds: a custom `baseUrl` imposes its own body limit.
+
 The routed-subscription registry axes (`upstream-rotation`, `region-upstreams-global|us|eu`, `region-limits-eu`, `credit-rates`, `list-price-from`, `routing-family`, `policy-aliases`, `entitlement`, `default-reasoning-off`) describe a gateway whose proxy fans one model out to several upstreams. They are read through `src/compat/factory-droid.ts` by Factory Droid discovery and its request provider, not materialized by `buildModel`. A provider-wide `region-upstreams-*` rule is the upstream serving table and a model rule replaces it for that region. `list-price-from "<provider>" ["<id>"]` shows a bundled row's list price beside the subscription's own billing; unlike seed values it is resolved at runtime and degrades to the seed's zero cost when the row is gone. See [`providers/factory-droid.kdl`](providers/factory-droid.kdl), whose wire and billing pool per model are `api-routes` and `quota-tiers` rules in `runtime/behavior.kdl`.
 
 ### Time-based pricing

package/src/compat/rules/classes/anthropic.kdl

@@ -72,7 +72,7 @@ class "anthropic" {
 			thinking-prefix-binding #true
 		}
 	}
-	// Adaptive Claude (Opus 4.7+, Sonnet/Fable/Mythos 5+) rejects explicit
+	// Adaptive Claude (Opus 4.7+, Sonnet/Fable/Mythos 5+, Haiku 5.5+) rejects explicit
 	// sampling params (`temperature`, `top_p`, `top_k`) with a 400 on every
 	// serving host. The `<10` bound keeps whole-number revisions such as
 	// `claude-opus-45` (Opus 4.5) out of the range.
@@ -96,6 +96,11 @@ class "anthropic" {
 			supports-sampling-params #false
 		}
 	}
+	family "haiku" {
+		revision ">=5.5 <10" {
+			supports-sampling-params #false
+		}
+	}
 	// On-demand compaction (compact-2026-09-04) works on the Claude API, Vertex,
 	// and Amazon Bedrock's `/anthropic` routes, not Bedrock Converse.
 	// Foundry / Claude Platform on AWS are first-party Anthropic routes whose
@@ -121,6 +126,11 @@ class "anthropic" {
 				supports-server-compaction #true
 			}
 		}
+		family "haiku" {
+			revision ">=5.5 <6" {
+				supports-server-compaction #true
+			}
+		}
 		models "claude-mythos-preview" {
 			supports-server-compaction #true
 		}
@@ -173,6 +183,15 @@ class "anthropic" {
 			thinking-prefix-binding #true
 		}
 	}
+	// Haiku 5.5 binds signed thinking to the exact preceding conversation like
+	// Sonnet 5.5. It accepts a forced `tool_choice` (the response just skips
+	// thinking), so it keeps forced tool selection.
+	// https://platform.claude.com/docs/en/models/haiku-5-5/migration-guide
+	family "haiku" {
+		revision ">=5.5 <6" {
+			thinking-prefix-binding #true
+		}
+	}
 	on "anthropic" "cloudflare-ai-gateway" "google-vertex" {
 		family "fable" {
 			revision ">=5" {
@@ -200,6 +219,13 @@ class "anthropic" {
 				supports-turn-scoped-system #true
 			}
 		}
+		family "haiku" {
+			revision ">=5.5 <6" {
+				supports-mid-conversation-system #true
+				supports-mid-conversation-tool-changes #true
+				supports-turn-scoped-system #true
+			}
+		}
 		family "opus" {
 			revision ">=4.8" {
 				supports-mid-conversation-system #true
@@ -230,6 +256,13 @@ class "anthropic" {
 				supports-per-message-effort #true
 			}
 		}
+		// Haiku 5.5 mirrors Sonnet 5.5; Vertex stays off until verified.
+		family "haiku" {
+			revision ">=5.5 <6" {
+				supports-thinking-binding-controls #true
+				supports-per-message-effort #true
+			}
+		}
 		family "opus" {
 			revision ">=5" {
 				supports-per-message-effort #true
@@ -257,6 +290,10 @@ class "anthropic" {
 		"google-vertex" "opencode-zen" "snowflake" "vercel-ai-gateway" {
 		family "haiku" {
 			thinking-mode "budget"
+			// Haiku 5.5 is the first adaptive Haiku: `budget_tokens` is a 400.
+			revision ">=5.5 <6" {
+				thinking-mode "anthropic-adaptive"
+			}
 		}
 	}
 	on "amazon-bedrock" "anthropic" "cloudflare-ai-gateway" "github-copilot" "gitlab-duo" \
@@ -365,6 +402,34 @@ class "anthropic" {
 				max-tokens 128000
 			}
 		}
+		// /v1/models reports neither reasoning, vision, pricing nor limits for
+		// Haiku 5.5, so a discovered row would materialize as a text-only,
+		// unpriced, non-reasoning model; pin the model card. Prompts over 100K
+		// input tokens bill the whole request at the long-context rates.
+		// https://platform.claude.com/docs/en/models/haiku-5-5/overview
+		family "haiku" {
+			revision ">=5.5 <6" {
+				thinking-upgrade-neutral #true
+				input-modalities "text" "image"
+				cost-patch {
+					input 0.1
+					output 0.5
+					cache-read 0.01
+					cache-write 0.125
+				}
+				long-context-cost {
+					input-threshold 100000
+					input 0.5
+					output 2.5
+					cache-read 0.05
+					cache-write 0.625
+				}
+				limits-patch {
+					context-window 1000000
+					max-tokens 128000
+				}
+			}
+		}
 		family "mythos" {
 			cost-patch {
 				input 10.0

package/src/compat/rules/providers/anthropic.kdl

@@ -34,6 +34,12 @@ provider "anthropic" {
 			cost input=2 output=10 cache-read=0.2 cache-write=2.5
 			limits context=1000000 max-tokens=128000
 		}
+		model "claude-haiku-5-5" name="Claude Haiku 5.5" {
+			reasoning #true
+			input "text" "image"
+			cost input=0.1 output=0.5 cache-read=0.01 cache-write=0.125
+			limits context=1000000 max-tokens=128000
+		}
 		model "claude-fable-5" name="Claude Fable 5" {
 			reasoning #true
 			input "text" "image"
@@ -68,6 +74,13 @@ provider "anthropic" {
 	// combines this fact with a per-request effective-URL check, so it never
 	// authorizes a block on its own.
 	first-party-provider #true
+	// Messages requests cap at 32 MB. Reserve 8 MB for text, tools, and JSON
+	// framing when retaining inline base64 images in a long conversation.
+	// Applied only on the official endpoint; a custom baseUrl has its own limit.
+	// https://platform.claude.com/docs/en/api/overview#request-size-limits
+	on-api "anthropic-messages" {
+		inline-image-byte-budget 24000000
+	}
 	class "anthropic" {
 		revision ">=3.7 <4.6" {
 			thinking-efforts "minimal" "low" "medium" "high" "xhigh"

package/src/compat/rules/providers/moonshot.kdl

@@ -31,6 +31,17 @@ provider "moonshot" {
 			input-modalities "text" "image"
 		}
 	}
+	// stencil.so reports a $3 K3 cache write; Moonshot bills only cache hits
+	// ($0.30) and misses (the $3 input rate), with no cache-write charge.
+	// https://platform.moonshot.ai/docs/pricing/chat-k3
+	models "kimi-k3" {
+		cost-patch {
+			input 3.0
+			output 15.0
+			cache-read 0.3
+			cache-write 0.0
+		}
+	}
 	// The /models roster carries no modality metadata for the named vision/vl
 	// SKUs; the bounded id token is the census signal.
 	models token="vision" {