[Feat] Add azure/gpt-5.5 + azure/gpt-5.5-pro entries (+ dated variants)

[mateo-berri]

Relevant issues

Day-0 Azure entries for OpenAI's GPT-5.5 family. Microsoft hasn't shipped GPT-5.5 on Azure OpenAI yet (latest GA on the Foundry models page is the GPT-5.4 series as of 2026-04-24), but the existing precedent — azure/gpt-5.4* was in the cost map before its Azure rollout — means we should land these entries now so cost tracking + capability flags Just Work the moment customers deploy.

This PR was force-pushed to expand scope from a single azure/gpt-5.5 entry to the full set of four entries (chat + pro, base + dated variants), match the long-context pricing structure of gpt-5.4, and add tests. The previous narrower revision (#26458) is now superseded and closed.

Pre-Submission checklist

• I have Added testing in the tests/test_litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem
• I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Screenshots / Proof of Fix

$ uv run pytest tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py \
                tests/test_litellm/test_utils.py::test_aaamodel_prices_and_context_window_json_is_valid \
                -k "azure_gpt55 or json_is_valid"
7 passed

Type

🆕 New Feature

Changes

Adds four entries to both model_prices_and_context_window.json and litellm/model_prices_and_context_window_backup.json:

Model	mode	input	output	cached	context
azure/gpt-5.5	chat	$5.00	$30.00	$0.50	1.05M
azure/gpt-5.5-2026-04-23	chat	$5.00	$30.00	$0.50	1.05M
azure/gpt-5.5-pro	responses	$60.00	$360.00	$6.00	1.05M
azure/gpt-5.5-pro-2026-04-23	responses	$60.00	$360.00	$6.00	1.05M

Pricing per-1M-tokens. Schema follows the existing azure/gpt-5.4* shape:

• Same base + long-context pricing as the openai/gpt-5.5* counterparts
• Azure-style: priority tier present (2x base), no flex / batches keys
• mode: chat for thinking, mode: responses for pro

reasoning_effort capability flags mirror the OpenAI variants since Azure proxies the same API contract — minimal rejected on both chat and pro, low and none rejected on pro.

Dependency note

supports_low_reasoning_effort is set on azure/gpt-5.5-pro* here, but the schema entry + transformation logic that reads the flag is introduced in #26456. Until #26456 lands, the flag on the Azure entries is inert (no effect — low continues to pass through). After #26456 merges, both OpenAI and Azure pro variants will reject low consistently.

The schema-test fix (supports_low_reasoning_effort registered in tests/test_litellm/test_utils.py) is included here so test_aaamodel_prices_and_context_window_json_is_valid passes regardless of merge order.

Tests

tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py:

• test_azure_gpt55_entries_present_with_correct_pricing — parametrized over all four entries; verifies provider, mode, base + long-context pricing, context window
• test_azure_gpt55_reasoning_effort_flags_match_live_openai_api — pins the live-API-derived supports_{none, minimal, xhigh}_reasoning_effort profile

tests/test_litellm/test_utils.py:

• Schema validator updated to accept supports_low_reasoning_effort alongside the existing reasoning-effort flags

[CLAassistant]

All committers have signed the CLA.

[veria-ai]

Low: No security issues

This PR adds Azure GPT-5.5 model entries to the static model cost/pricing JSON files and adds corresponding tests. The changes are purely data additions to JSON configuration files with no impact on authentication, authorization, input validation, or any runtime code paths.

---

Status: 0 open
Risk: 1/10

Posted by Veria AI · 2026-04-24T22:41:32.572Z

[codspeed-hq]

Merging this PR will not alter performance

✅ 16 untouched benchmarks

---

_{Comparing litellm_hotfix_gpt-5-5-azure (72649c7) with main (d21e90f)}

[greptile-apps]

Greptile Summary

This PR adds four Azure GPT-5.5 model entries (azure/gpt-5.5, azure/gpt-5.5-2026-04-23, azure/gpt-5.5-pro, azure/gpt-5.5-pro-2026-04-23) to the cost map, along with accompanying tests and a schema validator extension for supports_low_reasoning_effort.

• The two dated variants (azure/gpt-5.5-2026-04-23 and azure/gpt-5.5-pro-2026-04-23) are missing the reasoning-effort capability flags (supports_none_reasoning_effort, supports_xhigh_reasoning_effort, supports_minimal_reasoning_effort, and for the pro variant also supports_low_reasoning_effort) that their base entries declare. The OpenAI counterparts (gpt-5.5-2026-04-23) correctly include these fields, so the omission will cause get_model_info on the dated variants to silently return None for these flags.

Confidence Score: 4/5

Safe to merge after adding the missing reasoning-effort flags to the two dated Azure variants.

One P1 finding: the dated variants are missing reasoning-effort capability flags that their base entries declare, causing silent None returns from get_model_info. No security issues, no pricing errors, no test coverage regressions. The fix is mechanical (copy the three/four boolean fields into the dated variant objects).

model_prices_and_context_window.json and litellm/model_prices_and_context_window_backup.json — both need reasoning-effort flags added to the dated variant entries.

Important Files Changed

Filename	Overview
model_prices_and_context_window.json	Adds azure/gpt-5.5, azure/gpt-5.5-2026-04-23, azure/gpt-5.5-pro, azure/gpt-5.5-pro-2026-04-23 entries; dated variants are missing reasoning-effort capability flags present in the base entries and their OpenAI counterparts
litellm/model_prices_and_context_window_backup.json	Mirror of model_prices_and_context_window.json — same four Azure GPT-5.5 entries with the same gap in reasoning-effort flags on dated variants
tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py	Adds pricing and reasoning-effort flag tests for azure/gpt-5.5*; reasoning-effort test only covers base models, not dated variants, so the missing flags go undetected
tests/test_litellm/test_utils.py	Extends JSON schema validator to allow supports_low_reasoning_effort field — straightforward schema update

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[PR adds 4 Azure model entries] --> B["azure/gpt-5.5 (base)"]
    A --> C["azure/gpt-5.5-2026-04-23 (dated)"]
    A --> D["azure/gpt-5.5-pro (base)"]
    A --> E["azure/gpt-5.5-pro-2026-04-23 (dated)"]

    B --> B1["✅ supports_none_reasoning_effort: true"]
    B --> B2["✅ supports_xhigh_reasoning_effort: true"]
    B --> B3["✅ supports_minimal_reasoning_effort: false"]

    C --> C1["❌ supports_none_reasoning_effort: MISSING"]
    C --> C2["❌ supports_xhigh_reasoning_effort: MISSING"]
    C --> C3["❌ supports_minimal_reasoning_effort: MISSING"]

    D --> D1["✅ supports_none_reasoning_effort: false"]
    D --> D2["✅ supports_xhigh_reasoning_effort: true"]
    D --> D3["✅ supports_minimal_reasoning_effort: false"]
    D --> D4["✅ supports_low_reasoning_effort: false"]

    E --> E1["❌ supports_none_reasoning_effort: MISSING"]
    E --> E2["❌ supports_xhigh_reasoning_effort: MISSING"]
    E --> E3["❌ supports_minimal_reasoning_effort: MISSING"]
    E --> E4["❌ supports_low_reasoning_effort: MISSING"]

Loading

_{Reviews (2): Last reviewed commit: "test: register supports_low_reasoning_ef..." | Re-trigger Greptile}

[greptile-apps]

supports_web_search missing from parity fields

supports_web_search is not included in parity_fields, so if it is ever added to gpt-5.5 (OpenAI) but not backported to azure/gpt-5.5, this test will silently pass while the entries diverge. All comparable azure/gpt-4.1* entries carry "supports_web_search": false, suggesting it is expected to be declared explicitly.

Suggested change

	openai_entry = litellm.model_cost["gpt-5.5"]
	azure_entry = litellm.model_cost["azure/gpt-5.5"]
	# Provider differs by design; everything else should match.
	assert openai_entry["litellm_provider"] == "openai"
	assert azure_entry["litellm_provider"] == "azure"
	parity_fields = [
	"input_cost_per_token",
	"output_cost_per_token",
	"cache_read_input_token_cost",
	"max_input_tokens",
	"max_output_tokens",
	"max_tokens",
	"mode",
	"supported_endpoints",
	"supported_modalities",
	"supported_output_modalities",
	"supports_function_calling",
	"supports_native_streaming",
	"supports_parallel_function_calling",
	"supports_pdf_input",
	parity_fields = [
	"input_cost_per_token",
	"output_cost_per_token",
	"cache_read_input_token_cost",
	"max_input_tokens",
	"max_output_tokens",
	"max_tokens",
	"mode",
	"supported_endpoints",
	"supported_modalities",
	"supported_output_modalities",
	"supports_function_calling",
	"supports_native_streaming",
	"supports_parallel_function_calling",
	"supports_pdf_input",
	"supports_prompt_caching",
	"supports_reasoning",
	"supports_response_schema",
	"supports_system_messages",
	"supports_tool_choice",
	"supports_service_tier",
	"supports_vision",
	"supports_web_search",
	"supports_none_reasoning_effort",
	"supports_xhigh_reasoning_effort",
	"supports_minimal_reasoning_effort",
	]

[Michael-RZ-Berri]

Looks good probably add the web search thing + hit endpoint once

[greptile-apps]

Dated variant missing reasoning-effort flags

azure/gpt-5.5-2026-04-23 is missing supports_none_reasoning_effort, supports_xhigh_reasoning_effort, and supports_minimal_reasoning_effort that its base entry (azure/gpt-5.5) declares. The corresponding OpenAI dated snapshot gpt-5.5-2026-04-23 correctly includes all three flags. Any code that calls get_model_info("azure/gpt-5.5-2026-04-23") and reads these fields will silently receive None instead of True/False, misreporting the model's capabilities.

The same gap exists for azure/gpt-5.5-pro-2026-04-23 (missing supports_none_reasoning_effort, supports_xhigh_reasoning_effort, supports_minimal_reasoning_effort, and supports_low_reasoning_effort compared to azure/gpt-5.5-pro).

The flags should be added to both dated variants to match the base entry, mirroring how gpt-5.5-2026-04-23 mirrors gpt-5.5 on the OpenAI side.