Add OpenAI's latest GPT-5.4-mini and GPT-5.4-nano model

[PeterDaveHello]

Reference:

• https://developers.openai.com/api/docs/models/gpt-5.4-mini
• https://developers.openai.com/api/docs/models/gpt-5.4-nano
• https://openai.com/index/introducing-gpt-5-4-mini-and-nano/

[qodo-free-for-open-source-projects]

Review Summary by Qodo

Add support for GPT-5.4-mini and GPT-5.4-nano models

✨ Enhancement

Walkthroughs

Description

• Add GPT-5.4-mini model with 400K context window support
• Add GPT-5.4-nano model with 400K context window support
• Add unit tests for both new model variants
• Update GPT-5 model detection tests with new variants

Diagram

flowchart LR
  A["Model Registry"] -->|"Add gpt-5.4-mini"| B["400K Context Window"]
  A -->|"Add gpt-5.4-nano"| C["400K Context Window"]
  D["Test Suite"] -->|"Add unit tests"| E["Max tokens validation"]
  F["GPT-5 Detection"] -->|"Update model list"| G["New variants included"]

Loading

File Changes

1. pr_agent/algo/__init__.py ✨ Enhancement +2/-0

Register new GPT-5.4 mini and nano models

• Added gpt-5.4-mini model with 400K context window to model registry
• Added gpt-5.4-nano model with 400K context window to model registry
• Both entries include comments indicating 400K limit with potential config restrictions

pr_agent/algo/init.py

---

2. tests/unittest/test_get_max_tokens.py 🧪 Tests +26/-0

Add max tokens tests for new models

• Added test_gpt54_mini_model_max_tokens test method validating 400K token limit
• Added test_gpt54_nano_model_max_tokens test method validating 400K token limit
• Both tests follow existing GPT-5.4 test pattern with mocked settings

tests/unittest/test_get_max_tokens.py

---

3. tests/unittest/test_litellm_reasoning_effort.py 🧪 Tests +2/-0

Update GPT-5 model detection test coverage

• Added gpt-5.4-nano to GPT-5 model detection test list
• Added gpt-5.4-mini to GPT-5 model detection test list
• Ensures new variants are recognized in reasoning effort logic

tests/unittest/test_litellm_reasoning_effort.py

---

[qodo-free-for-open-source-projects]

Code Review by Qodo

🐞 Bugs (1) 📘 Rule violations (3) 📎 Requirement gaps (0)

1. Hardcoded gpt-5.4-mini tokens 📘 Rule violation ⛯ Reliability

Description

The PR adds hardcoded model max-token defaults (400000) in Python code instead of sourcing them
from Dynaconf TOML settings. This makes runtime behavior less adjustable and violates the repo’s
configuration management requirement.

Code

pr_agent/algo/init.py[R47-48]

+    'gpt-5.4-mini': 400000,  # 400K, but may be limited by config.max_model_tokens
+    'gpt-5.4-nano': 400000,  # 400K, but may be limited by config.max_model_tokens

Evidence

Compliance requires runtime behavior/configuration (including model/provider settings and
thresholds) to be adjustable via .pr_agent.toml or pr_agent/settings/*.toml, but the PR
introduces new hardcoded model token limits in code.

AGENTS.md
pr_agent/algo/init.py[47-48]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
New model max-token values for `gpt-5.4-mini` and `gpt-5.4-nano` are hardcoded in Python code, but this repo requires runtime behavior/configuration to be adjustable via `.pr_agent.toml` and/or `pr_agent/settings/*.toml`.
## Issue Context
Model/provider settings and thresholds should live in Dynaconf TOML defaults/overrides so they can be changed without code edits.
## Fix Focus Areas
- pr_agent/algo/__init__.py[47-48]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

2. Single quotes in new tests 📘 Rule violation ✓ Correctness

Description

The newly added unit tests introduce single-quoted strings in multiple places, diverging from the
repository preference for double quotes. This can trigger formatting/lint inconsistencies and reduce
style consistency.

Code

tests/unittest/test_get_max_tokens.py[R39-44]

+        fake_settings = type('', (), {
+            'config': type('', (), {
+                'custom_model_max_tokens': 0,
+                'max_model_tokens': 0
+            })()
+        })()

Evidence

The formatting rule explicitly prefers double quotes; the new test code uses single quotes for
strings (including type('', (), ...) and dict keys), which is a direct style deviation introduced
by this PR.

AGENTS.md
tests/unittest/test_get_max_tokens.py[39-44]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
New unit test code uses single quotes, but the repository formatting standard prefers double quotes.
## Issue Context
Keeping quoting consistent reduces lint churn and keeps diffs aligned with Ruff/formatter expectations.
## Fix Focus Areas
- tests/unittest/test_get_max_tokens.py[39-44]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

3. Duplicated fake_settings setup 📘 Rule violation ✓ Correctness

Description

The new tests duplicate the same fake_settings construction logic for both mini and nano cases.
This increases maintenance burden and violates the duplication-reduction guideline.

Code

tests/unittest/test_get_max_tokens.py[R39-56]

+        fake_settings = type('', (), {
+            'config': type('', (), {
+                'custom_model_max_tokens': 0,
+                'max_model_tokens': 0
+            })()
+        })()
+
+        monkeypatch.setattr(utils, "get_settings", lambda: fake_settings)
+
+        assert get_max_tokens("gpt-5.4-mini") == 400000
+
+    def test_gpt54_nano_model_max_tokens(self, monkeypatch):
+        fake_settings = type('', (), {
+            'config': type('', (), {
+                'custom_model_max_tokens': 0,
+                'max_model_tokens': 0
+            })()
+        })()

Evidence

Compliance requires factoring repeated logic into variables/helpers; both new tests repeat the same
nested type(...-based settings stub, which could be extracted to a helper/fixture or parameterized
test.

tests/unittest/test_get_max_tokens.py[39-56]
Best Practice: Learned patterns

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
Two newly added tests duplicate the same `fake_settings` construction and `monkeypatch.setattr(utils, "get_settings", ...)` setup.
## Issue Context
Reducing duplication makes future changes (e.g., adding more config fields) less error-prone and keeps tests consistent.
## Fix Focus Areas
- tests/unittest/test_get_max_tokens.py[39-56]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

4. Token limit rationale mismatch 🐞 Bug ✓ Correctness

Description

MAX_TOKENS sets gpt-5.4-mini/nano to 400000 while adjacent gpt-5.4 entries are explicitly documented
as a 272000 “safe default without opt-in,” but the new entries don’t explain why they differ.
Because get_max_tokens() directly feeds token-budget decisions (e.g., diff pruning), this
inconsistency makes it easy to misconfigure/overtrust these values later.

Code

pr_agent/algo/init.py[R47-48]

+    'gpt-5.4-mini': 400000,  # 400K, but may be limited by config.max_model_tokens
+    'gpt-5.4-nano': 400000,  # 400K, but may be limited by config.max_model_tokens

Evidence

The same MAX_TOKENS table documents gpt-5.4 as a deliberately conservative 272K “safe default,” yet
gpt-5.4-mini/nano are set to 400K with no similar rationale. get_max_tokens() returns these table
values (unless capped by config), and pr_processing uses get_max_tokens(model) to decide whether to
return the full diff or prune it—so these numbers are operationally significant and should be
internally consistent/justified.

pr_agent/algo/init.py[45-48]
pr_agent/algo/utils.py[991-1012]
pr_agent/algo/pr_processing.py[72-80]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
`MAX_TOKENS` documents `gpt-5.4` as a 272K **safe default** but assigns `gpt-5.4-mini`/`gpt-5.4-nano` a 400K limit without explaining the difference. These values are consumed by `get_max_tokens()` and directly influence prompt/diff budgeting decisions.
### Issue Context
This is primarily an internal consistency/operability issue: future maintainers (or config authors) can’t tell whether 400K is intentional, a different policy for mini/nano, or an oversight relative to the documented `gpt-5.4` safe-default approach.
### Fix Focus Areas
- pr_agent/algo/__init__.py[45-48]
- tests/unittest/test_get_max_tokens.py[25-60]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

ⓘ The new review experience is currently in Beta. Learn more

Previous review results

Review updated until commit 55d4c73

Results up to commit 901b784

🐞 Bugs (1) 📘 Rule violations (3) 📎 Requirement gaps (0)

1. Hardcoded gpt-5.4-mini tokens 📘 Rule violation ⛯ Reliability

Description

The PR adds hardcoded model max-token defaults (400000) in Python code instead of sourcing them
from Dynaconf TOML settings. This makes runtime behavior less adjustable and violates the repo’s
configuration management requirement.

Code

pr_agent/algo/init.py[R47-48]

+    'gpt-5.4-mini': 400000,  # 400K, but may be limited by config.max_model_tokens
+    'gpt-5.4-nano': 400000,  # 400K, but may be limited by config.max_model_tokens

Evidence

Compliance requires runtime behavior/configuration (including model/provider settings and
thresholds) to be adjustable via .pr_agent.toml or pr_agent/settings/*.toml, but the PR
introduces new hardcoded model token limits in code.

AGENTS.md
pr_agent/algo/init.py[47-48]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
New model max-token values for `gpt-5.4-mini` and `gpt-5.4-nano` are hardcoded in Python code, but this repo requires runtime behavior/configuration to be adjustable via `.pr_agent.toml` and/or `pr_agent/settings/*.toml`.

## Issue Context
Model/provider settings and thresholds should live in Dynaconf TOML defaults/overrides so they can be changed without code edits.

## Fix Focus Areas
- pr_agent/algo/__init__.py[47-48]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

2. Single quotes in new tests 📘 Rule violation ✓ Correctness

Description

The newly added unit tests introduce single-quoted strings in multiple places, diverging from the
repository preference for double quotes. This can trigger formatting/lint inconsistencies and reduce
style consistency.

Code

tests/unittest/test_get_max_tokens.py[R39-44]

+        fake_settings = type('', (), {
+            'config': type('', (), {
+                'custom_model_max_tokens': 0,
+                'max_model_tokens': 0
+            })()
+        })()

Evidence

The formatting rule explicitly prefers double quotes; the new test code uses single quotes for
strings (including type('', (), ...) and dict keys), which is a direct style deviation introduced
by this PR.

AGENTS.md
tests/unittest/test_get_max_tokens.py[39-44]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
New unit test code uses single quotes, but the repository formatting standard prefers double quotes.

## Issue Context
Keeping quoting consistent reduces lint churn and keeps diffs aligned with Ruff/formatter expectations.

## Fix Focus Areas
- tests/unittest/test_get_max_tokens.py[39-44]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

3. Duplicated fake_settings setup 📘 Rule violation ✓ Correctness

Description

The new tests duplicate the same fake_settings construction logic for both mini and nano cases.
This increases maintenance burden and violates the duplication-reduction guideline.

Code

tests/unittest/test_get_max_tokens.py[R39-56]

+        fake_settings = type('', (), {
+            'config': type('', (), {
+                'custom_model_max_tokens': 0,
+                'max_model_tokens': 0
+            })()
+        })()
+
+        monkeypatch.setattr(utils, "get_settings", lambda: fake_settings)
+
+        assert get_max_tokens("gpt-5.4-mini") == 400000
+
+    def test_gpt54_nano_model_max_tokens(self, monkeypatch):
+        fake_settings = type('', (), {
+            'config': type('', (), {
+                'custom_model_max_tokens': 0,
+                'max_model_tokens': 0
+            })()
+        })()

Evidence

Compliance requires factoring repeated logic into variables/helpers; both new tests repeat the same
nested type(...-based settings stub, which could be extracted to a helper/fixture or parameterized
test.

tests/unittest/test_get_max_tokens.py[39-56]
Best Practice: Learned patterns

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
Two newly added tests duplicate the same `fake_settings` construction and `monkeypatch.setattr(utils, "get_settings", ...)` setup.

## Issue Context
Reducing duplication makes future changes (e.g., adding more config fields) less error-prone and keeps tests consistent.

## Fix Focus Areas
- tests/unittest/test_get_max_tokens.py[39-56]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

4. Token limit rationale mismatch 🐞 Bug ✧ Quality

Description

MAX_TOKENS sets gpt-5.4-mini/nano to 400000 while adjacent gpt-5.4 entries are explicitly documented
as a 272000 “safe default without opt-in,” but the new entries don’t explain why they differ.
Because get_max_tokens() directly feeds token-budget decisions (e.g., diff pruning), this
inconsistency makes it easy to misconfigure/overtrust these values later.

Code

pr_agent/algo/init.py[R47-48]

+    'gpt-5.4-mini': 400000,  # 400K, but may be limited by config.max_model_tokens
+    'gpt-5.4-nano': 400000,  # 400K, but may be limited by config.max_model_tokens

Evidence

The same MAX_TOKENS table documents gpt-5.4 as a deliberately conservative 272K “safe default,” yet
gpt-5.4-mini/nano are set to 400K with no similar rationale. get_max_tokens() returns these table
values (unless capped by config), and pr_processing uses get_max_tokens(model) to decide whether to
return the full diff or prune it—so these numbers are operationally significant and should be
internally consistent/justified.

pr_agent/algo/init.py[45-48]
pr_agent/algo/utils.py[991-1012]
pr_agent/algo/pr_processing.py[72-80]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

### Issue description
`MAX_TOKENS` documents `gpt-5.4` as a 272K **safe default** but assigns `gpt-5.4-mini`/`gpt-5.4-nano` a 400K limit without explaining the difference. These values are consumed by `get_max_tokens()` and directly influence prompt/diff budgeting decisions.

### Issue Context
This is primarily an internal consistency/operability issue: future maintainers (or config authors) can’t tell whether 400K is intentional, a different policy for mini/nano, or an oversight relative to the documented `gpt-5.4` safe-default approach.

### Fix Focus Areas
- pr_agent/algo/__init__.py[45-48]
- tests/unittest/test_get_max_tokens.py[25-60]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

[qodo-free-for-open-source-projects]

Persistent review updated to latest commit 55d4c73