fix: added claude haiku support in backend rust

[shodown96]

Add claude-haiku-4-5-20251001 support to desktop Rust backend

The Swift desktop app sends claude-haiku-4-5-20251001 to /v2/chat/completions, but the model was absent from MODEL_ROUTES, causing resolve_model() to return None and the proxy to reject every request with a 400 Bad Request. The backend log shows:

chat_completions: unknown model 'claude-haiku-4-5-20251001' from user <uid>

Changes

• desktop/Backend-Rust/src/models/chat_completions.rs - added omi-haiku and claude-haiku-4-5-20251001 entries to MODEL_ROUTES
• desktop/Backend-Rust/src/routes/chat_completions.rs - added Haiku pricing to model_cost(), added a dedicated test_resolve_model_haiku test (matching the pattern of test_resolve_model_sonnet and test_resolve_model_opus), and removed omi-haiku from test_resolve_model_unknown where it didn't belong

Note on the test changes

test_resolve_model_unknown previously included assert!(resolve_model("omi-haiku").is_none()). Rather than flipping that assertion to is_some() (which would be semantically wrong in a test called unknown), I removed it from that test entirely and added a proper dedicated test with full public_model, upstream_model, and provider assertions. My read is the omi-haiku line was added there simply because Haiku wasn't implemented yet, if it was deliberately excluded, I'd be happy to revert.

[greptile-apps]

Greptile Summary

This PR fixes a 400 Bad Request error on /v2/chat/completions by registering claude-haiku-4-5-20251001 (and the omi-haiku alias) in MODEL_ROUTES and adding the corresponding per-token pricing to model_cost().

• models/chat_completions.rs — two new ModelRoute entries (omi-haiku → claude-haiku-4-5-20251001 and the direct pass-through); remaining diff is formatting-only.
• routes/chat_completions.rs — Haiku pricing added ($0.80/$4.00/$0.08/$1.00 per MTok for input/output/cache-read/cache-write); test_resolve_model_unknown updated to assert omi-haiku now resolves to Some, but the assertion is misplaced — it belongs in a dedicated known-model test with a stronger upstream-model check.

Confidence Score: 4/5

Safe to merge — the core fix (model routing and pricing) is correct and the only concern is a test assertion placed in the wrong test function.

The model route entries and pricing values look correct. The one rough edge is that the omi-haiku assertion was dropped into test_resolve_model_unknown rather than a dedicated test, and it does not verify the upstream_model value the way every other route test does — so a future regression would go undetected.

The test block in desktop/Backend-Rust/src/routes/chat_completions.rs around test_resolve_model_unknown deserves a second look to verify the assertion is moved and strengthened before merging.

Important Files Changed

Filename	Overview
desktop/Backend-Rust/src/models/chat_completions.rs	Adds omi-haiku and claude-haiku-4-5-20251001 entries to MODEL_ROUTES; remaining diff is pure formatting (brace-on-same-line and line-wrap cleanup) with no logic change.
desktop/Backend-Rust/src/routes/chat_completions.rs	Adds Haiku pricing to model_cost() (correct per-token values) and moves an omi-haiku assertion into the wrong test (test_resolve_model_unknown), where it contradicts the test's purpose.

Sequence Diagram

sequenceDiagram
    participant App as Swift Desktop App
    participant Proxy as Rust Proxy /v2/chat/completions
    participant Resolver as resolve_model()
    participant Cost as model_cost()
    participant Anthropic as Anthropic API

    App->>Proxy: POST /v2/chat/completions {model: "claude-haiku-4-5-20251001"}
    Proxy->>Resolver: resolve_model("claude-haiku-4-5-20251001")
    note over Resolver: Scans MODEL_ROUTES (new entry added by this PR)
    Resolver-->>Proxy: Some(ModelRoute { upstream: "claude-haiku-4-5-20251001" })
    Proxy->>Anthropic: Forward request with upstream model
    Anthropic-->>Proxy: Response + usage tokens
    Proxy->>Cost: model_cost("claude-haiku-4-5-20251001")
    note over Cost: $0.80/$4.00/$0.08/$1.00 per MTok
    Cost-->>Proxy: ModelCost
    Proxy-->>App: OpenAI-format response

Loading

Comments Outside Diff (1)

1. desktop/Backend-Rust/src/routes/chat_completions.rs, line 953-958 (link)

   The is_some() assertion for omi-haiku is misplaced inside test_resolve_model_unknown, which is designed exclusively to verify that unrecognised model strings return None. Placing a known-model assertion here makes the test name actively misleading. The haiku assertion should also mirror how other model tests are written — verifying the concrete upstream_model value rather than just checking that resolution succeeds. The existing test_resolve_model_known_models or a new dedicated test is a better home for this.

   Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

_{Reviews (1): Last reviewed commit: "fix: added claude haiku support in backe..." | Re-trigger Greptile}

[shodown96]

Updated the PR in response to Greptile's concerns

Changes

• desktop/Backend-Rust/src/models/chat_completions.rs - added omi-haiku and claude-haiku-4-5-20251001 entries to MODEL_ROUTES
• desktop/Backend-Rust/src/routes/chat_completions.rs - added Haiku pricing to model_cost(), added a dedicated test_resolve_model_haiku test (matching the pattern of test_resolve_model_sonnet and test_resolve_model_opus), and removed omi-haiku from test_resolve_model_unknown where it didn't belong