llm.rb 12.1.0 → 12.2.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: e4c7d2a7e0fd6a5b05cfdaaf2702f94f9a15d3e277739559f5cb676768fe6a9c
-  data.tar.gz: ad3081a38e1096459791ec6cffd1ae0a8550f5d4df2c6c95fd8842b78a0024f4
+  metadata.gz: 20d8c7c9c3d9d270f371781b28e7537a5c6921cff138cfc895bdbf0042003d07
+  data.tar.gz: 11b1abd397f1b13afadf4fe56e9670037a7c6ee586e945fd6a909505a4114122
 SHA512:
-  metadata.gz: 0e5a58f5b22d07b05d981423414de270b1ae760c7920452e736a55709c6c1eeee941407de8a8667365ee83943deeb0554d05bec5425b7132fe907dc62f349ffa
-  data.tar.gz: fabedc3aad36ba5cbaf7b126c55ddc3cbb428aadb3b2887a40e429daf35c65ff81a4d8c903583dea27c051d89f076d9d3446736e948954a6cef26065aebfa624
+  metadata.gz: d99ad697a3b3d5d13a0804ab3cea569ffbf5dc88d853aabe7461c4ae59020519e140aa0de1c2b26336b03998212410449da238ea8df5f6a190012f026093464c
+  data.tar.gz: dd6314a5e1e7ba3899d1ab13ba4f076508481bc2e24d3502bfa5cc4ebfb66a2203772ffc9acb4987b741eea413997493b8f85a2a7a841cf768e80d9b412886a8

data/CHANGELOG.md

@@ -15,8 +15,89 @@
 
 ## What's next
 
+Changes since `v12.2.0`.
+
+## v12.2.0
+
 Changes since `v12.1.0`.
 
+This release adds Mistral as a new provider with chat completions, streaming, tool calls,
+structured outputs, file/image attachments, and embeddings support. It introduces
+the `trace:` option to `LLM::Agent#repl` for keeping the tracer active during
+interactive sessions.
+
+Several fixes land for the Google provider (generationConfig parameter leakage),
+`LLM::Context#tracer=` (always assigning nil), `LLM::Provider#with_tracer(nil)`
+(nil fallback), and `LLM::Context#repair!` (dropping Struct returns).
+
+The default HTTP timeout has been increased from 60s to 180s to better accommodate
+reasoning models and large structured outputs, and the Anthropic default model has
+been updated to `claude-opus-4-8`. Model metadata has been refreshed across
+Anthropic, AWS Bedrock, DeepInfra, Google, and xAI, with Mistral model data added
+to the registry.
+
+### Add
+
+* **Add `trace:` option to `LLM::Agent#repl`** <br>
+  `LLM::Agent#repl` now accepts a `trace:` keyword argument. By default
+  the tracer is disabled for the duration of the repl session to prevent
+  curses UI interference from output written to `$stdout` or `$stderr`.
+  Set `trace: true` to keep the tracer active during the session, which
+  is useful when the tracer writes to a file rather than the terminal.
+
+* **Add a new provider: LLM::Mistral** <br>
+  [Mistral](https://mistral.ai) is now supported through its
+  OpenAI-compatible API. The provider supports chat completions,
+  streaming, tool calls, structured output (schema), file/image
+  attachments, and embeddings. Use `LLM.mistral(...)` to create a
+  provider instance.
+
+* **Add `LLM.mistral(...)` convenience method** <br>
+  A new top-level accessor (`LLM.mistral`) returns an `LLM::Mistral`
+  provider instance, matching the pattern used by other providers.
+
+### Fix
+
+* **Fix Google `generationConfig` parameter leakage** <br>
+  Fix a bug in the Google provider where non-generation parameters
+  (`role`, `model`, `messages`, `stream`) were leaking into the
+  `generationConfig` object alongside legitimate generation config
+  parameters such as `temperature`. Non-config parameters are now
+  filtered out before constructing `generationConfig`.
+
+* **Fix `LLM::Context#tracer=` always assigning `nil`** <br>
+  The `LLM::Context#tracer=` setter had a bug where it always assigned
+  `nil` regardless of the tracer value passed. It now correctly assigns
+  the given tracer or falls back to `LLM::Tracer::Null`.
+
+* **Fix `LLM::Provider#with_tracer(nil)` fallback** <br>
+  `LLM::Provider#with_tracer(nil)` now falls back to
+  `LLM::Tracer::Null` instead of setting a `nil` tracer directly.
+
+* **Fix `LLM::Context#repair!` dropping `Struct` returns** <br>
+  `LLM::Context#repair!` used `[*prompt]` to wrap the prompt before
+  grepping for return objects. Since `LLM::Function::Return` is a
+  `Struct`, the splat operator expanded it into its member values
+  instead of wrapping it, causing the grep to silently drop returns.
+  The fix wraps both sources in an array before flattening.
+
+### Change
+
+* **Increase default provider timeout from 60s to 180s** <br>
+  The default HTTP timeout for all providers has been increased from
+  60 to 180 seconds to better accommodate long-running requests such
+  as reasoning models and large structured outputs.
+
+* **Change Anthropic default model to `claude-opus-4-8`** <br>
+  The default Anthropic chat model has been updated from
+  `claude-sonnet-4-20250514` to `claude-opus-4-8`, reflecting the
+  latest model release from Anthropic.
+
+* **Refresh model metadata** <br>
+  Update model listings, pricing, and capabilities for Anthropic,
+  AWS Bedrock, DeepInfra, Google, and xAI. Add Mistral model data
+  to the registry.
+
 ## v12.1.0
 
 Changes since `v12.0.0`.

data/README.md

@@ -23,7 +23,7 @@ that are opt-in.
 ## Features
 
 The runtime supports OpenAI, OpenAI-compatible endpoints, Anthropic, Google
-Gemini, DeepSeek, DeepInfra, xAI, Z.ai, AWS Bedrock, Ollama, and llama.cpp.
+Gemini, Mistral, DeepSeek, DeepInfra, xAI, Z.ai, AWS Bedrock, Ollama, and llama.cpp.
 It has first-class support for streaming, tool calls,  MCP
 and A2A, embeddings, vector stores and the RAG pattern.
 
@@ -265,28 +265,37 @@ agent.talk "perform research"
 <summary>What providers does llm.rb support?</summary>
 <br>
 <p>
-China-based
 
-* DeepSeek
-* zAI
+**Cloud**
 
-US-based
+The following cloud-based providers are available to choose from. <br>
+In no particular order:
 
-* OpenAI
-* Google (Gemini)
-* xAI
-* AWS bedrock
-* DeepInfra
-* Anthropic
+🇺🇸 OpenAI <br>
+🇺🇸 DeepInfra <br>
+🇺🇸 xAI <br>
+🇺🇸 Google (Gemini) <br>
+🇺🇸 AWS bedrock <br>
+🇺🇸 Anthropic <br>
+🇨🇳 DeepSeek <br>
+🇨🇳 zAI <br>
+🇪🇺 Mistral <br>
 
-Openweights
+**Weights**
 
-* DeepSeek
-* zAI
-* DeepInfra
-* AWS bedrock
+The following providers provide access to open-weight models. <br>
+In no particular order:
 
-Host your own
+🇺🇸 DeepInfra <br>
+🇺🇸 AWS bedrock <br>
+🇨🇳 DeepSeek <br>
+🇨🇳 zAI <br>
+🇪🇺 Mistral <br>
+
+**Local**
+
+The following providers can be run locally on your own hardware. <br>
+In no particular order:
 
 * Ollama
 * Llamacpp

data/data/anthropic.json

@@ -31,7 +31,7 @@
       "tool_call": true,
       "structured_output": true,
       "temperature": true,
-      "knowledge": "2025-03-31",
+      "knowledge": "2025-05",
       "release_date": "2025-11-24",
       "last_updated": "2025-11-24",
       "modalities": {
@@ -321,7 +321,7 @@
       "tool_call": true,
       "structured_output": true,
       "temperature": true,
-      "knowledge": "2025-03-31",
+      "knowledge": "2025-05",
       "release_date": "2025-11-24",
       "last_updated": "2025-11-01",
       "modalities": {
@@ -368,6 +368,7 @@
       "tool_call": true,
       "structured_output": true,
       "temperature": false,
+      "knowledge": "2026-01",
       "release_date": "2026-05-28",
       "last_updated": "2026-05-28",
       "modalities": {

data/data/bedrock.json

@@ -92,6 +92,47 @@
         "cache_write": 3.75
       }
     },
+    "jp.anthropic.claude-haiku-4-5-20251001-v1:0": {
+      "id": "jp.anthropic.claude-haiku-4-5-20251001-v1:0",
+      "name": "Claude Haiku 4.5 (JP)",
+      "description": "Fast Claude model for responsive assistance, classification, and lightweight agents",
+      "family": "claude-haiku",
+      "attachment": true,
+      "reasoning": true,
+      "reasoning_options": [
+        {
+          "type": "budget_tokens",
+          "min": 1024
+        }
+      ],
+      "tool_call": true,
+      "structured_output": true,
+      "temperature": true,
+      "knowledge": "2025-02-28",
+      "release_date": "2025-10-15",
+      "last_updated": "2025-10-15",
+      "modalities": {
+        "input": [
+          "text",
+          "image",
+          "pdf"
+        ],
+        "output": [
+          "text"
+        ]
+      },
+      "open_weights": false,
+      "limit": {
+        "context": 200000,
+        "output": 64000
+      },
+      "cost": {
+        "input": 1,
+        "output": 5,
+        "cache_read": 0.1,
+        "cache_write": 1.25
+      }
+    },
     "us.meta.llama4-scout-17b-instruct-v1:0": {
       "id": "us.meta.llama4-scout-17b-instruct-v1:0",
       "name": "Llama 4 Scout 17B Instruct (US)",
@@ -1349,6 +1390,7 @@
       ],
       "tool_call": true,
       "temperature": false,
+      "knowledge": "2026-01",
       "release_date": "2026-05-28",
       "last_updated": "2026-05-28",
       "modalities": {
@@ -2328,6 +2370,7 @@
       ],
       "tool_call": true,
       "temperature": false,
+      "knowledge": "2026-01",
       "release_date": "2026-05-28",
       "last_updated": "2026-05-28",
       "modalities": {
@@ -2835,6 +2878,7 @@
       ],
       "tool_call": true,
       "temperature": false,
+      "knowledge": "2026-01",
       "release_date": "2026-05-28",
       "last_updated": "2026-05-28",
       "modalities": {
@@ -3136,6 +3180,7 @@
       ],
       "tool_call": true,
       "temperature": false,
+      "knowledge": "2026-01",
       "release_date": "2026-05-28",
       "last_updated": "2026-05-28",
       "modalities": {
@@ -3181,6 +3226,7 @@
       ],
       "tool_call": true,
       "temperature": false,
+      "knowledge": "2026-01",
       "release_date": "2026-05-28",
       "last_updated": "2026-05-28",
       "modalities": {
@@ -3518,6 +3564,7 @@
       ],
       "tool_call": true,
       "temperature": false,
+      "knowledge": "2026-01",
       "release_date": "2026-05-28",
       "last_updated": "2026-05-28",
       "modalities": {

data/data/deepinfra.json

@@ -557,7 +557,7 @@
       },
       "open_weights": true,
       "limit": {
-        "context": 16384,
+        "context": 262144,
         "output": 65536
       },
       "cost": {
@@ -767,7 +767,7 @@
         "output": 16384
       },
       "cost": {
-        "input": 0.039,
+        "input": 0.037,
         "output": 0.17
       }
     },

data/data/google.json

@@ -562,6 +562,38 @@
         "input_audio": 0.3
       }
     },
+    "gemini-omni-flash-preview": {
+      "id": "gemini-omni-flash-preview",
+      "name": "Gemini Omni Flash Preview",
+      "description": "Video generation and editing model for fast, conversational text- and image-to-video workflows",
+      "family": "gemini",
+      "attachment": true,
+      "reasoning": true,
+      "reasoning_options": [],
+      "tool_call": false,
+      "temperature": true,
+      "release_date": "2026-06-30",
+      "last_updated": "2026-06-30",
+      "modalities": {
+        "input": [
+          "text",
+          "image",
+          "video"
+        ],
+        "output": [
+          "video"
+        ]
+      },
+      "open_weights": false,
+      "limit": {
+        "context": 131072,
+        "output": 65536
+      },
+      "cost": {
+        "input": 1.5,
+        "output": 17.5
+      }
+    },
     "gemini-3.1-flash-image-preview": {
       "id": "gemini-3.1-flash-image-preview",
       "name": "Nano Banana 2",