Expert and integration roles on claude-haiku-5-5 #134

2026-10-08T16:43:40Z

dev-agent commented

2026-10-08 16:43:40 +00:00

claude-haiku-4-5 is being replaced by claude-haiku-5-5 in LiteLLM. Both keys now see the new model, with its window and output limits declared (1M in, 128k out) and tool calls supported.

Changes

tests/integration/nisse-test.yaml: the integration tests' driver, helper and expert move to claude-haiku-5-5.
nisse-dev.example.yaml: the expert, and the capability profile, move to the new name. The profile's benchmark figure was measured on 4.5 and is labelled as such until it's re-measured.
The plan's "models by role today" line, which also now lists the local classifier (M3.2).
Left as they were: benchmark results, the Tuning Guide's measurement table, and the M3.2 eval note. All were measured on 4.5.
Outside the repo: the bring-up config on this box has its expert on claude-haiku-5-5, with the old file kept as nisse-dev.yaml.pre-haiku-5-5.

Checked locally on Haiku 5.5:

Integration tier: 18 passed. Haiku 5.5 was every role there, and that includes the live benchmark task.
Capability matrix: 7 of 7.
Classifier eval set: 9 of 9.

Once merged, claude-haiku-4-5 and the unused local models can come out of LiteLLM. Nothing live refers to them: the remaining references are made-up names in unit tests' fake catalogs, docs examples, and benchmark history.

🤖 Generated with Claude Code

https://claude.ai/code/session_01PPQ14L3LikyefCMFBqEbKE

`claude-haiku-4-5` is being replaced by **`claude-haiku-5-5`** in LiteLLM. Both keys now see the new model, with its window and output limits declared (1M in, 128k out) and tool calls supported. **Changes** - **`tests/integration/nisse-test.yaml`:** the integration tests' driver, helper and expert move to `claude-haiku-5-5`. - **`nisse-dev.example.yaml`:** the expert, and the capability profile, move to the new name. The profile's benchmark figure was measured on 4.5 and is labelled as such until it's re-measured. - **The plan's "models by role today" line,** which also now lists the local classifier (M3.2). - **Left as they were:** benchmark results, the Tuning Guide's measurement table, and the M3.2 eval note. All were measured on 4.5. - **Outside the repo:** the bring-up config on this box has its expert on `claude-haiku-5-5`, with the old file kept as `nisse-dev.yaml.pre-haiku-5-5`. **Checked locally on Haiku 5.5:** - **Integration tier:** 18 passed. Haiku 5.5 was every role there, and that includes the live benchmark task. - **Capability matrix:** 7 of 7. - **Classifier eval set:** 9 of 9. **Once merged**, `claude-haiku-4-5` and the unused local models can come out of LiteLLM. Nothing live refers to them: the remaining references are made-up names in unit tests' fake catalogs, docs examples, and benchmark history. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_01PPQ14L3LikyefCMFBqEbKE

dev-agent added 1 commit 2026-10-08 16:43:40 +00:00

Expert and integration roles on claude-haiku-5-5

CI / checks (push) Skipped

CI / checks (pull_request) Successful in 2m6s

Details

CI / e2e (pull_request) Successful in 3m58s

Details

5e1086c501

claude-haiku-4-5 is being replaced by claude-haiku-5-5 in LiteLLM: the
integration tests' roles and the example config's expert move to it.
Its profile keeps the predecessor's benchmark figures, marked as such
until it's measured. The plan's models-by-role line also gains the
local classifier (M3.2). Results measured on 4.5 stay as they were.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPQ14L3LikyefCMFBqEbKE

dongar approved these changes

2026-10-08 16:44:41 +00:00

dev-agent merged commit 33f9a75bac into main

2026-10-08 16:50:15 +00:00

dev-agent deleted branch haiku-5-5

2026-10-08 16:50:15 +00:00

Sign in to join this conversation.

2 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: dev/nisse#134