mirror of
https://github.com/docker/docker-agent.git
synced 2026-11-04 10:02:15 +00:00
Arnaud Héritier
a0c951ee4a
- examples/vercel.yaml: correct openai/gpt-4.1 (which moved backward from the pre-sweep openai/gpt-5) to openai/gpt-5.6-sol, matching every other vercel+openai reference touched by this PR (DefaultModels['vercel'], docs/providers/vercel, docs/providers/custom). - examples/custom_provider.yaml: bump the two leftover shorthand model refs (my_gateway/gpt-4o-mini, my_anthropic/claude-sonnet-4-5) that were missed in the same file, and rename gateway_gpt4o -> gateway_gpt so the key no longer encodes a stale model generation. - Rename model keys that encoded a now-stale generation in their own name after only the value was bumped: examples/lifecycle.yaml (openai/gpt-4o-mini -> openai/gpt-5.6-terra, updating the agent reference too), examples/alloy.yaml (gpt-4o -> gpt), examples/rag.yaml (gpt-5-minimal -> gpt-minimal), examples/multi-code.yaml (gpt4o -> gpt), examples/image_text_extractor.yaml (gpt4_vision -> gpt_vision), examples/thinking_budget.yaml (claude-4-5-sonnet-* -> claude-4-6-sonnet-*, claude-opus-4-6-* -> claude-opus-4-8-*, claude-opus-4-7-summarized -> claude-opus-4-8-summarized, including the cross-referencing comment), docs/providers/google/index.md (gemini-3-pro/gemini-3-flash -> gemini-pro/gemini-flash). - Fix minor markdown table column misalignment in docs/providers/baseten and docs/providers/github-copilot introduced by the previous commit. Refs #4137.
90 lines
3.2 KiB
YAML
90 lines
3.2 KiB
YAML
# Run the demo command with:
|
|
# docker agent run thinking_budget.yaml -c demo
|
|
|
|
agents:
|
|
root:
|
|
model: gpt-5-mini-min # <- try with gpt-5-mini-high, gpt-5-6-max, or gpt-5-6-none
|
|
# model: claude-4-6-sonnet-min # <- try with claude-4-6-sonnet-high or claude-opus-4-8-adaptive
|
|
# model: gemini-2-5-flash-dynamic-thinking # <- try with -no-thinking, -low or -high variants
|
|
description: a helpful assistant that thinks
|
|
instruction: you are a helpful assistant who can also use tools, but only if you need to
|
|
commands:
|
|
demo: "hey i need python code for a mandelbrot fractal"
|
|
toolsets:
|
|
- type: shell
|
|
|
|
models:
|
|
gpt-5-mini-min:
|
|
provider: openai
|
|
model: gpt-5-mini
|
|
thinking_budget: minimal # <- openai supports "none", "minimal", "low", "medium", "high", "xhigh", "max" (xhigh needs gpt-5.2+; none/max need gpt-5.6+; minimal is dropped on gpt-5.6+)
|
|
|
|
gpt-5-mini-high:
|
|
provider: openai
|
|
model: gpt-5-mini
|
|
thinking_budget: high
|
|
|
|
gpt-5-6-max:
|
|
provider: openai
|
|
model: gpt-5.6
|
|
thinking_budget: max # <- gpt-5.6+ only; the top-most OpenAI effort tier
|
|
|
|
gpt-5-6-none:
|
|
provider: openai
|
|
model: gpt-5.6
|
|
thinking_budget: none # <- gpt-5.6+ has a real "none" effort, honored on the wire instead of just disabling thinking locally
|
|
|
|
claude-4-6-sonnet-min:
|
|
provider: anthropic
|
|
model: claude-sonnet-4-6
|
|
thinking_budget: 1024 # <- explicit token budget (1024-32768) for older models
|
|
|
|
claude-4-6-sonnet-high:
|
|
provider: anthropic
|
|
model: claude-sonnet-4-6
|
|
thinking_budget: 32768 # <- explicit token budget (32768 is the Anthropic suggested maximum)
|
|
provider_opts:
|
|
interleaved_thinking: true # <- enables interleaved thinking, aka tool calling during model reasoning
|
|
|
|
claude-opus-4-8-adaptive:
|
|
provider: anthropic
|
|
model: claude-opus-4-8
|
|
thinking_budget: adaptive # <- lets the model decide when and how much to think (recommended for 4.6+)
|
|
|
|
claude-opus-4-8-low:
|
|
provider: anthropic
|
|
model: claude-opus-4-8
|
|
thinking_budget: low # <- adaptive thinking with low effort: "low", "medium", "high", "xhigh", "max" (same as adaptive/<effort>)
|
|
|
|
claude-opus-4-8-summarized:
|
|
provider: anthropic
|
|
model: claude-opus-4-8 # <- Opus 4.7 hides thinking by default; use the same flag with any recent Claude model
|
|
thinking_budget: adaptive
|
|
provider_opts:
|
|
thinking_display: summarized # <- "summarized" or "omitted" on 4.6+ models; "display" is pre-4.6 only (Opus 4.7 defaults to omitted)
|
|
|
|
gemini-2-5-flash-dynamic-thinking:
|
|
provider: google
|
|
model: gemini-2.5-flash
|
|
thinking_budget: -1 # <- google only, dynamic thinking
|
|
|
|
gemini-2-5-flash-no-thinking:
|
|
provider: google
|
|
model: gemini-2.5-flash
|
|
thinking_budget: 0 # <- google only, no thinking
|
|
|
|
gemini-2-5-flash-low:
|
|
provider: google
|
|
model: gemini-2.5-flash
|
|
thinking_budget: 1024
|
|
|
|
gemini-2-5-flash-high:
|
|
provider: google
|
|
model: gemini-2.5-flash
|
|
thinking_budget: 24576 # <- google's maximum thinking budget for all models except Gemini 2.5 Pro (max 32768)
|
|
|
|
bedrock-opus-4-8-adaptive:
|
|
provider: amazon-bedrock
|
|
model: global.anthropic.claude-opus-4-8
|
|
thinking_budget: adaptive/high # <- Opus 4.6+ reject token budgets on Bedrock too; token values are auto-converted to adaptive
|