llm.rb 16.0.0.pre.4 → 16.0.0.pre.6

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: a83847d3c33a48980bd9740d7aff397db757ec91224440a97c10240c0ae3a290
-  data.tar.gz: 402d6ea4a68fc987e3597d1b6dd84089f2804c7774ef35e38fb8fc424b0e62b0
+  metadata.gz: 0f5c0f9cb6f92099cec46e8c7a83e02498d375458a50b8b3f19a94ac2385dc8f
+  data.tar.gz: 11929258bad683f946ded02b7b2621fcfae30cf1ed11c57007e4b4cfbac05b5d
 SHA512:
-  metadata.gz: f75dda6e725c6aa48748a407e744c94c006d3db4973cf45064f7f0252fc2bc818ee7a99b564e66c4bad6ad4066bf5214b912a637b3b25b8b9f335d45d226c7e2
-  data.tar.gz: ea3e7fe4a32687c7a34f4b85a0bd4aeee5b7b68387452dd1ac98a3572f1b1f74b4f19ac3ee32b10cf6f53563b9eceb5a8e9292e6001b90728e4997c7b9a301a2
+  metadata.gz: 5b32ee2a9301958487a808705f63895b1cd89898730e8b0bdcabcee06be62eef6a724795ccd0f6053227f0577380019436eab3b19ba3a15459560bcc78a505af
+  data.tar.gz: 7839665a301ee3ab68716737daaddc7f384308b6b24548efcb33699a3463de83efa638aa060293a27ced6145907dee6ddabb7b63d11d5cb5f1e1a0239b2e209f

data/CHANGELOG.md

@@ -56,6 +56,29 @@
   has not been saved, where `#messages` reads the column. The Sequel plugin
   answers to `#messages!` too, where it is the same call as `#messages`.
 
+### Agent
+
+* **agent: keep the instructions it injected in step** <br>
+  [`LLM::Agent`](https://r.uby.dev/api-docs/llm.rb/LLM/Agent.html) now brings
+  the instructions it injected up to date before a request goes out, so a
+  conversation restored from saved state runs on the agent's current
+  instructions rather than on the ones it was saved with. The message is
+  found by a mark the agent puts on it, so a message the caller composed is
+  never replaced, and the mark is written by
+  [`LLM::Message#to_h`](https://r.uby.dev/api-docs/llm.rb/LLM/Message.html#to_h-instance_method)
+  and read back on a restore, which is the conversation the refresh exists
+  for. Before this, the instructions were injected once and the message
+  stayed as it was for the life of the conversation.
+
+* **agent: identify its own instructions by the mark, not by the role** <br>
+  Fix a bug where the agent decided whether it had already injected its
+  instructions by looking for a message with the `system` role. Google gives
+  instructions the `user` role, so the agent did not recognise the message it
+  had already injected and added another copy of the instructions on every
+  turn. It now asks the provider which role instructions are given, and
+  identifies the message it wrote by its mark, because on Google the message
+  a caller seeds a conversation with is a user message too.
+
 ### Console
 
 * **console: erase with the backspace key on OpenBSD** <br>
@@ -142,6 +165,18 @@
   once, when the whole turn had finished, so an interrupted turn persisted
   nothing.
 
+### Prompt
+
+* **prompt: let a message carry fields a caller attaches to it** <br>
+  [`LLM::Prompt#system`](https://r.uby.dev/api-docs/llm.rb/LLM/Prompt.html#system-instance_method),
+  [`#user`](https://r.uby.dev/api-docs/llm.rb/LLM/Prompt.html#user-instance_method),
+  [`#developer`](https://r.uby.dev/api-docs/llm.rb/LLM/Prompt.html#developer-instance_method),
+  and [`#talk`](https://r.uby.dev/api-docs/llm.rb/LLM/Prompt.html#talk-instance_method),
+  along with its `#chat` alias, now take an `extra:` keyword, so a message a
+  caller composes can carry fields of its own. The runtime reads those fields
+  back rather than inferring provenance from a role, which is how an agent
+  tells the instructions it injected from a message the caller wrote.
+
 ### Provider
 
 * **deepseek: support image attachments in chat completions** <br>
@@ -161,14 +196,17 @@
   Update `data/` with current model listings, limits, and pricing. Alibaba adds
   `qwen3.5-flash`, `qwen3.7-flash`, and `qwen3.8-omni-flash`, and gives
   `qwen3.7-plus` attachments, structured output, and a lower price; OpenAI adds
-  `gpt-daybreak-blue-latest` and `gpt-daybreak-red-latest`; Bedrock adds
-  `openai.gpt-6-sol`, `openai.gpt-6-luna`, and
-  `global.anthropic.claude-sonnet-5-5`; and DeepInfra adds `tencent/Hy4-preview`
-  and the Xiaomi MiMo V2.6 Pro and V2.6 Flash models.
+  `gpt-daybreak-blue-latest`, `gpt-daybreak-red-latest`, and `gpt-6.1-sol`, and
+  gives `gpt-6-astra` an `ultrafast` mode; Bedrock adds `openai.gpt-6-sol`,
+  `openai.gpt-6-luna`, `global.anthropic.claude-sonnet-5-5`,
+  `anthropic.claude-sonnet-5-5`, and the Grok 4.7 entries in both the global
+  and US regions; DeepInfra adds `tencent/Hy4-preview` and the Xiaomi MiMo
+  V2.6 Pro and V2.6 Flash models, and marks four older entries deprecated.
   Bedrock also corrects the context and output limits of fifteen models, Google
   lowers the limits of `gemini-2.5-computer-use-preview-10-2025` and
-  `gemini-3-pro-image`, and OpenRouter adds five models, drops six, and reprices
-  many of the DeepSeek, Z.ai, and Qwen entries.
+  `gemini-3-pro-image`, and OpenRouter adds seven models, `openai/gpt-6.1-sol`
+  and `openai/gpt-6.1-sol-pro` among them, drops six, and reprices or corrects
+  the limits of many of the DeepSeek, Z.ai, Qwen, and Tencent entries.
 
 ### Schema
 
@@ -195,6 +233,42 @@
   rate limit or a timeout calls it when it lands and not once per attempt, and
   it fires even when the stream is disabled.
 
+### Tracers
+
+* **tracer: report a request that failed** <br>
+  [`LLM::Tracer#on_request_error`](https://r.uby.dev/api-docs/llm.rb/LLM/Tracer.html#on_request_error-instance_method)
+  is now called wherever a request ends without a response, so a span that
+  [`LLM::Tracer#on_request_start`](https://r.uby.dev/api-docs/llm.rb/LLM/Tracer.html#on_request_start-instance_method)
+  returned is closed. Before this, a request was reported only when a provider
+  turned an error response into an exception, so a dropped connection or a
+  socket read that failed left the tracer holding a span it had opened and never
+  closed, which in a trace reads the same as a process that died. A failure that
+  surfaces as one of the transport's own error classes is reported the same way
+  unless the request was interrupted, and the exception the caller receives is
+  unchanged.
+
+* **tracer: add `on_interrupt`, called when a request is interrupted** <br>
+  [`LLM::Tracer#on_interrupt`](https://r.uby.dev/api-docs/llm.rb/LLM/Tracer.html#on_interrupt-instance_method)
+  is a new hook, called with a `scope:` of `:request` or `:tool` before
+  `LLM::Interrupt` reaches the caller, so a tracer that has to record what
+  happened to a turn does it while the work is still in flight. An interrupt is
+  the third ending a scope can have, beside `on_request_finish` and
+  `on_request_error`, and it is not reported as a failure. Unlike the other
+  request and tool hooks it does nothing by default, because an interrupt is
+  delivered to whatever tracer happens to be bound, and a hook that raised would
+  replace the interrupt every caller is written against. A request is announced
+  with the span that `on_request_start` returned and its `request_id`; a tool
+  pass has neither, so both arrive as `nil`.
+
+* **tracer: announce an interrupted tool phase once** <br>
+  [`LLM::Context#wait`](https://r.uby.dev/api-docs/llm.rb/LLM/Context.html#wait-instance_method)
+  now calls the hook once when an interrupt unwinds through it, rather than once
+  per tool, because a cancel reaches every tool that is running and the caller
+  hears a single exception, so the announcement belongs to the phase rather than
+  to a tool. The hook runs before the caller is given the interrupt. A `:ractor`
+  call that answers `cancelled: true` instead of raising is not announced,
+  because the phase was not interrupted; one call was retired.
+
 ## v15.5.0
 
 Changes since `v15.4.1`.

data/README.md

@@ -78,7 +78,8 @@ agent.talk "hello world"
 Streams can be simple IO objects or subclasses of
 [`LLM::Stream`](https://r.uby.dev/api-docs/llm.rb/LLM/Stream.html)
 with structured callbacks for content,
-reasoning, tool calls, tool returns, and compaction.
+reasoning, tool calls, tool returns, the boundary a completed
+request marks, and compaction.
 Streams can also observe message transformers, which rewrite
 outgoing messages before they reach the provider.
 
@@ -102,6 +103,11 @@ class MyStream < LLM::Stream
   def on_tool_return(tool, result)
   end
 
+  # A request has completed: the response is in the conversation, and
+  # any tools it asked for run after this.
+  def on_step(ctx, res)
+  end
+
   # Before a transformer rewrites an outgoing message.
   def on_transform(transformer)
   end

data/data/bedrock.json

@@ -777,6 +777,54 @@
       },
       "canonical_model_id": "anthropic/claude-opus-5-5"
     },
+    "anthropic.claude-sonnet-5-5": {
+      "id": "anthropic.claude-sonnet-5-5",
+      "name": "Claude Sonnet 5.5",
+      "description": "Fast Claude model for everyday coding, agents, and knowledge work",
+      "family": "claude-sonnet",
+      "attachment": true,
+      "reasoning": true,
+      "reasoning_options": [
+        {
+          "type": "effort",
+          "values": [
+            "low",
+            "medium",
+            "high",
+            "xhigh",
+            "max"
+          ]
+        }
+      ],
+      "tool_call": true,
+      "structured_output": false,
+      "temperature": false,
+      "knowledge": "2026-06",
+      "release_date": "2026-09-28",
+      "last_updated": "2026-09-28",
+      "modalities": {
+        "input": [
+          "text",
+          "image",
+          "pdf"
+        ],
+        "output": [
+          "text"
+        ]
+      },
+      "open_weights": false,
+      "limit": {
+        "context": 1000000,
+        "output": 128000
+      },
+      "cost": {
+        "input": 2,
+        "output": 10,
+        "cache_read": 0.2,
+        "cache_write": 2.5
+      },
+      "canonical_model_id": "anthropic/claude-sonnet-5-5"
+    },
     "global.anthropic.claude-sonnet-5-5": {
       "id": "global.anthropic.claude-sonnet-5-5",
       "name": "Claude Sonnet 5.5 (Global)",
@@ -3717,6 +3765,52 @@
       },
       "canonical_model_id": "anthropic/claude-opus-5"
     },
+    "global.xai.grok-4.7": {
+      "id": "global.xai.grok-4.7",
+      "name": "Grok 4.7 (Global)",
+      "description": "xAI's frontier model for long-running agents, coding, knowledge work, and visual projects",
+      "family": "grok",
+      "attachment": true,
+      "reasoning": true,
+      "reasoning_options": [
+        {
+          "type": "effort",
+          "values": [
+            "low",
+            "medium",
+            "high",
+            "xhigh"
+          ]
+        }
+      ],
+      "tool_call": true,
+      "structured_output": true,
+      "temperature": true,
+      "knowledge": "2026-05",
+      "release_date": "2026-09-21",
+      "last_updated": "2026-09-28",
+      "modalities": {
+        "input": [
+          "text",
+          "image",
+          "pdf"
+        ],
+        "output": [
+          "text"
+        ]
+      },
+      "open_weights": false,
+      "limit": {
+        "context": 500000,
+        "output": 500000
+      },
+      "cost": {
+        "input": 2,
+        "output": 6,
+        "cache_read": 0.5
+      },
+      "canonical_model_id": "xai/grok-4.7"
+    },
     "us.openai.gpt-5.6-sol": {
       "id": "us.openai.gpt-5.6-sol",
       "name": "GPT-5.6 Sol (US)",
@@ -4229,6 +4323,52 @@
       },
       "canonical_model_id": "mistral/ministral-3-3b-instruct-2512"
     },
+    "us.xai.grok-4.7": {
+      "id": "us.xai.grok-4.7",
+      "name": "Grok 4.7 (US)",
+      "description": "xAI's frontier model for long-running agents, coding, knowledge work, and visual projects",
+      "family": "grok",
+      "attachment": true,
+      "reasoning": true,
+      "reasoning_options": [
+        {
+          "type": "effort",
+          "values": [
+            "low",
+            "medium",
+            "high",
+            "xhigh"
+          ]
+        }
+      ],
+      "tool_call": true,
+      "structured_output": true,
+      "temperature": true,
+      "knowledge": "2026-05",
+      "release_date": "2026-09-21",
+      "last_updated": "2026-09-28",
+      "modalities": {
+        "input": [
+          "text",
+          "image",
+          "pdf"
+        ],
+        "output": [
+          "text"
+        ]
+      },
+      "open_weights": false,
+      "limit": {
+        "context": 500000,
+        "output": 500000
+      },
+      "cost": {
+        "input": 2.2,
+        "output": 6.6,
+        "cache_read": 0.55
+      },
+      "canonical_model_id": "xai/grok-4.7"
+    },
     "us.openai.gpt-6-luna": {
       "id": "us.openai.gpt-6-luna",
       "name": "GPT-6 Luna (US)",

data/data/deepinfra.json

@@ -246,6 +246,7 @@
         "context": 1048576,
         "output": 16384
       },
+      "status": "deprecated",
       "cost": {
         "input": 1,
         "output": 3,
@@ -330,6 +331,7 @@
         "context": 262144,
         "output": 16384
       },
+      "status": "deprecated",
       "cost": {
         "input": 0.14,
         "output": 0.28,
@@ -1712,6 +1714,7 @@
         "context": 163840,
         "output": 163840
       },
+      "status": "deprecated",
       "cost": {
         "input": 0.24,
         "output": 0.9,
@@ -2022,6 +2025,7 @@
         "context": 262144,
         "output": 262144
       },
+      "status": "deprecated",
       "cost": {
         "input": 0.68,
         "output": 3.4,

data/data/openai.json

@@ -1121,6 +1121,99 @@
         "cache_read": 0.175
       }
     },
+    "gpt-6.1-sol": {
+      "id": "gpt-6.1-sol",
+      "name": "GPT-6.1 Sol",
+      "description": "Near-Astra performance for complex coding, computer use, and professional work at a lower cost.",
+      "family": "gpt-sol",
+      "attachment": true,
+      "reasoning": true,
+      "reasoning_options": [
+        {
+          "type": "effort",
+          "values": [
+            "low",
+            "medium",
+            "high",
+            "xhigh",
+            "max"
+          ]
+        }
+      ],
+      "tool_call": true,
+      "structured_output": true,
+      "temperature": false,
+      "knowledge": "2026-04-30",
+      "release_date": "2026-09-29",
+      "last_updated": "2026-09-29",
+      "modalities": {
+        "input": [
+          "text",
+          "image",
+          "pdf"
+        ],
+        "output": [
+          "text"
+        ]
+      },
+      "open_weights": false,
+      "limit": {
+        "context": 1050000,
+        "input": 922000,
+        "output": 128000
+      },
+      "experimental": {
+        "modes": {
+          "fast": {
+            "cost": {
+              "input": 4,
+              "output": 20,
+              "cache_read": 0.2,
+              "cache_write": 5
+            },
+            "provider": {
+              "body": {
+                "service_tier": "priority"
+              }
+            }
+          },
+          "pro": {
+            "provider": {
+              "body": {
+                "reasoning": {
+                  "mode": "pro"
+                }
+              }
+            }
+          }
+        }
+      },
+      "cost": {
+        "input": 2,
+        "output": 10,
+        "cache_read": 0.1,
+        "cache_write": 2.5,
+        "tiers": [
+          {
+            "input": 4,
+            "output": 15,
+            "cache_read": 0.2,
+            "cache_write": 5,
+            "tier": {
+              "type": "context",
+              "size": 272000
+            }
+          }
+        ],
+        "context_over_200k": {
+          "input": 4,
+          "output": 15,
+          "cache_read": 0.2,
+          "cache_write": 5
+        }
+      },
+      "canonical_model_id": "openai/gpt-6.1-sol"
+    },
     "text-embedding-3-large": {
       "id": "text-embedding-3-large",
       "name": "text-embedding-3-large",
@@ -1248,6 +1341,19 @@
                 }
               }
             }
+          },
+          "ultrafast": {
+            "cost": {
+              "input": 60,
+              "output": 300,
+              "cache_read": 6,
+              "cache_write": 75
+            },
+            "provider": {
+              "body": {
+                "service_tier": "ultrafast"
+              }
+            }
           }
         }
       },