fix(pricing): remove gpt-image-2 text output cost

[gorquan]

Relevant issues

Related to #31142, closed unmerged #26865, and open #27064

PR #26865 already noted that GPT Image 2 has no text output rate, but it was closed without merging after #26644 added the current entries. PR #27064 applies the same omission to separate azure_ai/... aliases and does not update these existing OpenAI and Azure entries

Linear ticket

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have added meaningful tests
• My PR passes all CI/CD checks (e.g., lint, format, unit tests)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Delays in PR merge?

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

Screenshots / Proof of Fix

No UI change

OpenAI lists GPT Image 2 text output as - while image output remains $30 per 1M tokens, and the model page lists image as its only output modality. Microsoft likewise states there is no billing for GPT Image 2 text output tokens

Sources: OpenAI pricing, OpenAI model, Microsoft Foundry announcement

Before at bf02a4a47f08ad8007a24c8da2ffa8de1a1d36a2:

gpt-image-2: text=0.00001, image=0.00003
gpt-image-2-2026-04-21: text=0.00001, image=0.00003
azure/gpt-image-2: text=0.00001, image=0.00003
azure/gpt-image-2-2026-04-21: text=0.00001, image=0.00003

After at 89f54feccafb10c990b90d60a3271ce0185bdc77:

gpt-image-2: text=absent, image=0.00003
gpt-image-2-2026-04-21: text=absent, image=0.00003
azure/gpt-image-2: text=absent, image=0.00003
azure/gpt-image-2-2026-04-21: text=absent, image=0.00003

Type

Bug Fix

Changes

Remove the unsupported output_cost_per_token field from the canonical and dated GPT Image 2 entries for OpenAI and Azure in both cost maps

Update model metadata and cost calculator regression tests so image output remains billed at output_cost_per_image_token and raw cost maps cannot reintroduce a text output rate

[CLAassistant]

All committers have signed the CLA.

[greptile-apps]

Greptile Summary

Removes the erroneous output_cost_per_token field from the four gpt-image-2 entries (canonical + Azure variants) in both cost-map files, aligning with OpenAI and Microsoft's published pricing where text output is not billed. Tests are updated to reflect the corrected expectation.

• The JSON deletions are precise and consistent across both model_prices_and_context_window.json and the backup; output_cost_per_image_token (3e-5) is preserved.
• The cost-calculator fallback in _get_token_base_cost (lines 222–227 of utils.py) already handles the case where output_cost_per_token is 0 by substituting output_cost_per_image_token, so the renamed "uses_image_rate" test correctly describes the resulting behavior.
• A new raw-JSON regression test (test_gpt_image_2_cost_maps_have_no_text_output_price) ensures neither cost map can silently reintroduce a text-output price for these models.

Confidence Score: 5/5

Safe to merge — the change removes a billing field that should never have existed; no logic is altered beyond the cost-map data.

The deletions are narrow, factually correct per upstream pricing documentation, and mirrored consistently across both JSON files. The _get_token_base_cost fallback (output_cost_per_image_token when output_cost_per_token is 0) was already in place, so the cost calculator behaves correctly without any code changes. The updated and newly added tests accurately describe the post-fix behavior and would catch any re-introduction of the text-output price.

No files require special attention.

Important Files Changed

Filename	Overview
model_prices_and_context_window.json	Removes incorrect output_cost_per_token from gpt-image-2 and gpt-image-2-2026-04-21 (both OpenAI and azure/ entries); image output rate preserved at 3e-5
litellm/model_prices_and_context_window_backup.json	Mirror of the main cost map change — same four entries corrected; backup stays in sync with canonical file
tests/test_litellm/test_utils.py	Updates three existing assertions from output_cost_per_token == 1e-05 to == 0, adds a raw-JSON regression test; the == 0 assertion is valid because get_model_info defaults the missing field to 0
tests/test_litellm/test_gpt_image_cost_calculator.py	Removes text-token parameters from breakdown test and updates fallback-path test to expect image rate (3e-5); both changes are consistent with the _get_token_base_cost fallback that uses output_cost_per_image_token when output_cost_per_token is 0

_{Reviews (1): Last reviewed commit: "fix(pricing): remove gpt-image-2 text ou..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing gorquan:fix-gpt-image-2-text-output-price (9acf527) with litellm_internal_staging (3e9e520)}

[devin-ai-integration]

Superseded: the gpt-image-2 output_cost_per_token removal is already on litellm_internal_staging. Registry work continues in rolling PR #31884.