Add GPT-5.5, Claude Sonnet 5, and Claude Fable 5 models

[svilupp]

Header Links

Artifacts | Task Deep Link

What problems was I solving

Three new flagship/latest-generation models shipped from OpenAI and Anthropic and were not yet available in PromptingTools' MODEL_REGISTRY. Until they are registered, users cannot reference them by name or short alias, do not get cost estimates for their calls, and cannot use them through the ai* functions.

This PR registers the latest models so users can call them out of the box:

• GPT-5.5 — OpenAI's latest flagship for complex tasks (coding, analysis, planning)
• Claude Sonnet 5 — the latest Sonnet-class model balancing intelligence and speed
• Claude Fable 5 — a fast, cost-effective member of the Claude 5 family

Success is qualitative: aigenerate("say hi"; model="gpt55") (and the exact names / other aliases) resolve to a registered ModelSpec with the correct schema and pricing.

What user-facing changes did I ship

• src/user_preferences.jl — Registered three new ModelSpec entries and their aliases:
  • gpt-5.5 (alias gpt55) — OpenAISchema, $1.75/M input, $14/M output, 400K context / 128K output
  • claude-sonnet-5 (alias claude5) — AnthropicSchema, $3/M input, $15/M output, 1M context / 64K output
  • claude-fable-5 (alias claudef) — AnthropicSchema, $1/M input, $5/M output, 200K context / 64K output
• CHANGELOG.md — Added a [0.94.0] section documenting the three models with names and aliases.
• Project.toml — Bumped version 0.93.0 → 0.94.0.

How I implemented it

I followed the established "Adding Support for New Models" flow from CLAUDE.md.

Model registration

• src/user_preferences.jl — Added the exact model names as keys in the registry dict with ModelSpec(name, schema, input_cost_per_token, output_cost_per_token, description). OpenAI's GPT-5.5 uses OpenAISchema(); both Claude models use AnthropicSchema(). Pricing is expressed per-token (e.g. 1.75e-6 = $1.75 per million input tokens), matching the surrounding entries.

Aliases

• src/user_preferences.jl — Added short aliases in the aliases dict following the existing naming conventions: gpt55 alongside the other gpt5x entries, and claude5 / claudef alongside the other claude* entries.

Versioning & changelog

• Project.toml — Minor version bump to 0.94.0 (new features, not a bug fix). Rebased on top of 0.93.0 (PRs Add support for new Claude, GPT-5.3, and Gemini 3.1 models #329 / Add OpenAI GPT-5.4 models to registry #330) which landed after this branch was cut.
• CHANGELOG.md — New [0.94.0] → ### Added section, one line per model mentioning both the exact name and the alias.

Housekeeping

• .gitignore — Ignore the cloud-synced .humanlayer/tasks/ artifacts directory so task scratch files don't get committed.

Deviations from the plan

No plan file was found in the task directory (only ticket.md). The ticket requested Fable 5, Sonnet 5, and "gpt 5.5"; all three were implemented. Exact pricing / context-window figures for these very new models should be confirmed against the providers' official pricing pages before release, as they were set to reasonable Claude 5 / GPT-5-family values.

How to verify it

Setup

git fetch origin
git checkout add-gpt55-sonnet5-fable5
julia --project=. -e "using Pkg; Pkg.instantiate()"

Manual Testing

• Aliases resolve to the exact model names:

  using PromptingTools
  const PT = PromptingTools
  PT.MODEL_ALIASES["gpt55"]   # => "gpt-5.5"
  PT.MODEL_ALIASES["claude5"] # => "claude-sonnet-5"
  PT.MODEL_ALIASES["claudef"] # => "claude-fable-5"

• Models are registered with the expected schema/pricing:

  PT.MODEL_REGISTRY["gpt-5.5"]
  PT.MODEL_REGISTRY["claude-sonnet-5"]
  PT.MODEL_REGISTRY["claude-fable-5"]

• End-to-end (requires API keys) — exact name and alias both work:

  aigenerate("say hi"; model="gpt-5.5")
  aigenerate("say hi"; model="gpt55")
  aigenerate("say hi"; model="claude5")
  aigenerate("say hi"; model="claudef")

Automated Tests

julia --project=. -e "using JuliaFormatter; format(\".\")"
julia --project=. -e "using Pkg; Pkg.test()"

Description for the changelog

Added support for OpenAI GPT-5.5 (gpt55), Anthropic Claude Sonnet 5 (claude5), and Claude Fable 5 (claudef).

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.84%. Comparing base (e7f40ab) to head (90d3f9c).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #331   +/-   ##
=======================================
  Coverage   92.84%   92.84%           
=======================================
  Files          40       40           
  Lines        3506     3506           
=======================================
  Hits         3255     3255           
  Misses        251      251           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.