langchain-zerogpu 1.3.0

Model catalog sync. The ZeroGPU API no longer serves glm-5.2, so
ZeroGPUReasonLongContextTool — which failed on every call — is gone, and the
GLiNER models behind the structured-extraction and PII tools have moved to the
GLiNER2.5 generation. Five models the API has added since the last sync now
have tools: a multimodal Qwen reasoner, OpenAI's GPT-6 Luna, a cheaper 3B
summarizer, and two decision models for routing work. Tool names and outputs
are unchanged apart from the notes below.

Added

• ZeroGPUReasonQwenTool (zerogpu_reason_qwen) calls qwen3.6-35b-a3b, a
  fast 35B mixture-of-experts model for document, screenshot, and content
  understanding, with structured output support over a 16K-token context — the
  smallest context of the reasoning tools.
• ZeroGPUReasonGPT6LunaTool (zerogpu_reason_gpt6_luna) calls gpt-6-luna,
  the most efficient model of OpenAI's GPT-6 family, with adjustable reasoning
  effort, function calling, and structured outputs over a 272K-token context.
  It costs half as much per input token as ZeroGPUReasonGPTLunaTool at the
  same context window.
• ZeroGPUSummarize3BTool (zerogpu_summarize_3b) calls
  llama-3.2-3b-instruct-fast, which shares llama-3.1-8b-instruct-fast's
  summarization pipeline and 131K-token context at a third of its input price.
• ZeroGPUDecideTool (zerogpu_decide) calls gliner2.5-decide-v1, a
  zero-shot decision model for intent classification, agent and tool routing,
  triage, moderation decisions, and human handoff, answering several decisions
  in one call against options supplied at request time. It is the cheapest
  model in the package per input token.
• ZeroGPUDecideToolsTool (zerogpu_decide_tools) calls
  zlm-v1-tool-decisions, which picks the tool a request needs and generates
  the query or arguments to call it with in a single inference step, returning
  NO_TOOL when no tool is needed.

Changed

• ZeroGPUClassifyStructuredTool, ZeroGPUExtractEntitiesTool, and
  ZeroGPUExtractJSONTool now call gliner2.5-base-v1. The ZeroGPU API no
  longer serves gliner2-base-v1, and gliner2.5-base-v1 is the schema-based
  extraction model it serves in its place.
• ZeroGPUExtractPIITool and ZeroGPURedactPIITool now call
  gliner2.5-multi-v1, the multilingual GLiNER2.5 extraction model. The API no
  longer serves gliner-multi-pii-v1.
• ZeroGPUReasonDeepseekTool: the cost comparison in the tool description
  agents route on no longer held. deepseek-v4.1-flash costs about three times
  as much per input token and three and a half times as much per output token
  as glm-5.3-flash — not 40% and 60% more — so the preference for
  ZeroGPUReasonGLMTool at the same 1M-token context is now a wide one.
• ZeroGPUReasonGPTMiniTool: its description named the removed
  ZeroGPUReasonLongContextTool as the only pricier tool. gpt-4.1-mini is
  now the priciest model in the package per token, and the description says so.
• ZeroGPUReasonGPTLunaTool: its description now points agents at the cheaper
  ZeroGPUReasonGPT6LunaTool, which has the same 272K-token context.

Removed

• ZeroGPUReasonLongContextTool (zerogpu_reason_long_context). The ZeroGPU
  API no longer serves glm-5.2, so the tool failed on every call. Remove it
  from your imports; ZeroGPUReasonGLMTool handles very large inputs at a
  1M-token context. With the five tools added above, the toolkit now returns
  twenty-six tools instead of twenty-two.

Install

pip install -U langchain-zerogpu==1.3.0

langchain-zerogpu 1.2.0

Model catalog sync. The ZeroGPU API no longer serves deepseek-v4-flash-0731,
so ZeroGPUReasonCodeTool — which failed on every call — is gone, and the four
text-generation models the API has added since the last sync now have tools:
glm-5.3-flash picks up the coding and agentic work at a 1M-token context,
alongside three OpenAI models. Tool names and outputs are unchanged apart from
the notes below.

Added

• ZeroGPUReasonGLMTool (zerogpu_reason_glm) calls glm-5.3-flash, Z.ai's
  efficient open-weight model for coding and long-horizon agent tasks, whose
  hybrid sparse and linear attention keeps long-context behaviour accurate
  across a 1M-token window, with function calling and adjustable reasoning
  effort. It is the cheapest tool with a 1M-token context.
• ZeroGPUReasonGPTLunaTool (zerogpu_reason_gpt_luna) calls gpt-5.6-luna,
  the cost-optimized model of OpenAI's GPT-5.6 family, with adjustable
  reasoning effort, function calling, and structured outputs over a 272K-token
  context.
• ZeroGPUReasonGPTMiniTool (zerogpu_reason_gpt_mini) calls gpt-4.1-mini,
  OpenAI's fast, cost-efficient GPT-4.1 model for instruction following and
  tool calling across a 1M-token context.
• ZeroGPUReasonGPTNanoTool (zerogpu_reason_gpt_nano) calls gpt-5.4-nano,
  the most cost-efficient model of OpenAI's GPT-5.4 family, built for
  high-volume, latency-sensitive classification, extraction, routing, and
  sub-agent work over a 400K-token context.

Changed

• ZeroGPUReasonDeepseekTool: the cost comparison in the tool description
  agents route on pointed at the removed ZeroGPUReasonCodeTool. It now points
  at ZeroGPUReasonGLMTool, the tool that keeps the same 1M-token context —
  deepseek-v4.1-flash costs about 40% more per input token and 60% more per
  output token than glm-5.3-flash.

Removed

• ZeroGPUReasonCodeTool (zerogpu_reason_code). The ZeroGPU API no longer
  serves deepseek-v4-flash-0731, so the tool failed on every call. Remove it
  from your imports; ZeroGPUReasonGLMTool covers the same coding and agentic
  work at the same 1M-token context. The toolkit now returns twenty-two tools
  instead of nineteen.

Install

pip install -U langchain-zerogpu==1.2.0

langchain-zerogpu 1.1.0

Model catalog sync. Three models the ZeroGPU API now serves had no tool, so the
toolkit returns nineteen tools instead of sixteen: a second DeepSeek Flash
reasoning tool, a Llama Guard moderation tool, and a signal-extraction tool for
topics, keywords, and intent. Because deepseek-v4.1-flash matches
deepseek-v4-flash-0731's 1M-token context window, ZeroGPUReasonCodeTool no
longer claims that window outright. Tool names and outputs are unchanged apart
from the notes below.

Added

• ZeroGPUReasonDeepseekTool (zerogpu_reason_deepseek) calls
  deepseek-v4.1-flash, a sparse mixture-of-experts model with a 1M-token
  context window, function calling, and both fast and higher-effort reasoning
  modes. It costs about twice as much per input token as
  ZeroGPUReasonCodeTool, which has the same context window.
• ZeroGPUModerateTool (zerogpu_moderate) calls llama-guard-4-12b (12B
  parameters, 160K-token context) and returns a safe / unsafe verdict plus the
  policy categories a violation matched — for chat moderation, prompt and
  response filtering, policy enforcement, and agent guardrails.
• ZeroGPUExtractSignalsTool (zerogpu_extract_signals) calls
  zlm-v1-signal-extract (80M parameters), which turns a passage into
  structured contextual signals — topics, keywords, intent — in one call, for
  content enrichment, ad targeting, agent routing, and analytics pipelines.
• The toolkit now returns nineteen tools instead of sixteen.

Changed

• ZeroGPUReasonCodeTool: deepseek-v4-flash-0731 no longer has the largest
  context window on the platform on its own — deepseek-v4.1-flash matches its
  1M tokens. The tool description agents route on now says the two tie on
  context and that this tool is the cheaper of them, instead of pointing at a
  single long-context alternative.

Install

pip install -U langchain-zerogpu==1.1.0

langchain-zerogpu 1.0.0

Model catalog sync. ZeroGPUFollowUpQuestionsTool is gone — the ZeroGPU API no
longer serves the model behind it, so every call it made was already failing.
ZeroGPUReasonCodeTool follows its model to a version-pinned id, and the
long-context tool had been advertising a context window four times larger than
glm-5.2 actually has, which was steering agents toward the most expensive
model on the platform for inputs that never needed it. Tool names and outputs
are unchanged apart from the notes below.

Changed

• ZeroGPUReasonCodeTool now calls deepseek-v4-flash-0731. The API renamed
  deepseek-v4-flash, and the old id no longer resolves. The model's 284B
  parameters and 1,048,576-token context window are unchanged.
• ZeroGPUReasonLongContextTool: glm-5.2's context window is 262K, not 1M.
  Corrected in the tool description agents route on, its docstring, and the
  README. It remains the most capable and priciest ZeroGPU model, but at roughly
  six to seven times zerogpu_reason per token rather than twenty. Because
  deepseek-v4-flash-0731 now has the largest context window in the catalog,
  ZeroGPUReasonCodeTool no longer defers to this tool on sheer input size.
• ZeroGPUReasonTool: gpt-oss-120b has 120B parameters, not 117B.
• ZeroGPUReasonMultilingualTool: qwen3-30b-a3b-fp8 has 30B parameters, not
  30.5B.

Removed

• ZeroGPUFollowUpQuestionsTool (zerogpu_followup_questions).
  zlm-v1-followup-questions-edge is no longer served by the ZeroGPU API, so
  the tool failed on every call. Remove it from your imports; the toolkit now
  returns sixteen tools instead of seventeen.

Install

pip install -U langchain-zerogpu==1.0.0

langchain-zerogpu 0.2.4

Catches the package up with the ZeroGPU API spec. Four models published in the
API reference had no tool — the two large open-weight reasoning models, the
follow-up question generator, and the domain-level IAB classifier — and the
enriched IAB classifier had moved to a v2 model id that the package was not
sending. All four are now first-class tools and the stale id is fixed.

The catalog then gained two more open-weight text-generation models, glm-5.2
and deepseek-v4-flash, both with a 1,048,576-token context window. Those are
wrapped too, taking the toolkit from eleven tools to seventeen.

The spec also made x-project-id optional, so the project id is no longer
required here either. Testing the new models against the live API turned up two
bugs on the existing surface: summarization had stopped working entirely, and
reasoning models would have returned their scratchpad instead of their answer.
Both are fixed.

Added

• ZeroGPUReasonTool (gpt-oss-120b) — a 117B open-weight model with a 131K
  context window, for prompts that need more reasoning headroom than the nano
  edge models.
• ZeroGPUReasonMultilingualTool (qwen3-30b-a3b-fp8) — reasoning across
  100+ languages. Served on the Chat Completions endpoint only, so it does not
  route through the Responses API.
• ZeroGPUReasonLongContextTool (glm-5.2) — a 753B MoE model with a
  1,048,576-token context, for inputs that do not fit zerogpu_reason's 131K:
  whole repositories, book-length documents, long agent transcripts. The most
  capable model on the platform and by a wide margin the most expensive —
  $1.10 / $3.50 per 1M input/output tokens, roughly twenty times
  gpt-oss-120b — so the tool description steers agents away from it unless
  the input genuinely requires the context. Chat Completions only.
• ZeroGPUReasonCodeTool (deepseek-v4-flash) — a 284B MoE model (13B active
  per token) with the same 1M context, tuned for coding and agentic workflows,
  at $0.07 / $0.14 per 1M. Chat Completions only. Note that this model is
  published in the API reference but not yet served: it currently returns
  404 model_not_found. The tool is correct per the spec and will work once
  the platform enables it.
• ZeroGPUFollowUpQuestionsTool (zlm-v1-followup-questions-edge) — returns
  the questions a reader would naturally ask next, as a list[str] (the model
  emits a JSON array; newline-delimited output is accepted as a fallback).
• ZeroGPUClassifyDomainTool (zlm-v1-iab-domain-classifier) — maps a bare
  domain name (e.g. nytimes.com) to IAB categories, topics, keywords, and
  user intent without fetching the page. Takes a domain argument rather than
  text, via the new DomainInput schema.
• ZeroGPUClient.chat() / ZeroGPUClient.achat(), reinstating the Chat
  Completions path (removed in 0.2.0) because qwen3-30b-a3b-fp8 is not served
  on /v1/responses, and llama-3.1-8b-instruct-fast is no longer reachable
  there. Like the Responses path, it recovers the reply from the response body
  when the SDK raises ParsingError on a 2xx.
• tests/unit_tests/test_models.py, pinning every tool to the model id listed
  in the API spec's model enum so a server-side rename fails in CI rather
  than at request time.
• HTTP 402 Payment Required from the ZeroGPU API now maps to a clear
  ZeroGPUError ("payment required … check your plan and billing details")
  instead of the generic "request failed" message.

Changed

• The project id is now optional, matching the API spec, which documents
  x-project-id as an optional header that scopes a request to a project.
  Constructing a tool or toolkit without ZEROGPU_PROJECT_ID no longer raises;
  requests simply go out unscoped. The API key is still required.

Fixed

• ZeroGPUSummarizeTool works again. llama-3.1-8b-instruct-fast rejects the
  Responses API's plain-string input with invalid_prompt ("required
  properties at '/' are 'prompt' ... 'messages'"), so every summarize call was
  failing; it now goes through Chat Completions, where that model is served. It
  also sends a short summarize instruction, without which the model replies
  conversationally to the passage and truncates mid-sentence at the completion
  cap.
• ZeroGPUClient no longer returns a reasoning model's scratchpad in place of
  its answer. gpt-oss-120b prepends a reasoning output item whose block
  also carries a text field, and the old "first text block wins" extraction
  would have returned that trace; the output_text block is now selected by
  type, with the previous behaviour kept as a fallback for models that do not
  tag their blocks.
• ZeroGPUClassifyIABEnrichedTool now calls
  zlm-v2-iab-classify-edge-enriched, the id published in the API spec. The
  package was still sending zlm-v1-iab-classify-edge-enriched, which the API
  currently still serves as an alias but no longer documents.
• ZeroGPUClassifyIABEnrichedTool's docstring still named the v1 id after
  the constant had moved to v2, so the rendered API documentation contradicted
  what the tool actually sent.
• tests/integration_tests/test_compile.py asserted a hard-coded tool count of
  eleven and had been failing since the count moved to fifteen. CI runs only the
  unit tests, so nothing caught it. It now tracks ALL_TOOL_CLASSES.

Install

pip install -U langchain-zerogpu==0.2.4

langchain-zerogpu 0.2.3

Bug-fix release. Every tool call failed against the current ZeroGPU API
because of a response-schema mismatch, and chat tools rejected a system
prompt. Both are fixed; no public API changes.

Fixed

• Tool calls no longer fail when the API response shape drifts from the
  zerogpu-api SDK's model (the server returns created_at while the pinned
  SDK expects created). ZeroGPUClient now recovers the output text from the
  response body when the SDK raises ParsingError on an otherwise successful
  (2xx) response, so all tools work again out of the box.
• The system argument on ZeroGPUChatTool and ZeroGPUChatThinkingTool no
  longer triggers a 400 invalid_request_error. The passage is sent as a
  plain-string input and the system prompt is carried as the Responses API
  instructions field, instead of a system/user message list that the
  endpoint rejects.

Install

pip install -U langchain-zerogpu==0.2.3

langchain-zerogpu 0.2.2

Documentation-only release: adds the ZeroGPU logo to the README so it renders
on the PyPI project page. The library is unchanged since 0.2.1.

Changed

• README.md now opens with the ZeroGPU logo (assets/logo.png, referenced
  by absolute URL so PyPI renders it).

Install

pip install -U langchain-zerogpu==0.2.2

langchain-zerogpu 0.2.1

First release published to PyPI. The library is unchanged since 0.2.0 — this
release establishes the automated release pipeline: tagged versions are built,
published to PyPI via Trusted Publishing (OIDC), and turned into GitHub
releases whose notes come straight from this changelog.

Added

• CI changelog job that fails pull requests changing package code without a
  matching CHANGELOG.md update.

Changed

• The Release workflow now creates the GitHub release from the matching
  ## [X.Y.Z] section of this changelog (titled langchain-zerogpu X.Y.Z,
  with an install block and the built artifacts attached) instead of
  auto-generated notes, and publishes to PyPI via Trusted Publishing.

Install

pip install -U langchain-zerogpu==0.2.1