Claude Haiku 5.5 on iac-bench 2026-10

Anthropic's Claude Haiku 5.5 scored an overall composite of 68.0 (95% CI 66.1–90.6) on iac-bench 2026-10, ranking 9th of 10 models evaluated on generating OpenTofu infrastructure modules, with 2 of 9 attempts failing the gate and 1 of 9 flagged as insecure. The low rank was driven by the Cloud SQL fixture, where two of three modules failed because the model looked up the environment's network inside a required project_id that its own example inputs set to the placeholder "example-dbops-project"; the model saw the tofu_plan error and submitted anyway, while its third attempt avoided the lookup and scored 0.929. Claude Haiku 5.5 tied Claude Opus 5.5 for the field-best median of 88 on the S3 fixture and placed fourth on App Service, at a list price of $0.10/$0.50 per million input/output tokens and an estimated $0.11 for all 9 attempts.

Upcoming Workshop: Citizen Developer Agentic Software Factory on AWS 11.4.26 @ 12PM PST Register Here https://streamyard.com/watch/umMggUFkVMNu How Claude Haiku 5.5 did on iac-bench 2026-10: an overall composite of 68.0, scores per fixture, strengths, weaknesses and the patterns in its OpenTofu modules. | Developer | Anthropic | | API identifier | claude-haiku-5-5 | | Access | Anthropic API | | Weights | Closed | | List price USD / M tokens, in / out | 0.10 / 0.50 OpenRouter, 2026-10-23 | | Evaluated | 2026-10-22 generation · 2026-10-23 judge | | Attempts | 9 3 fixtures × 3 samples | | Overall composite | 68.0 95% CI 66.1–90.6 | | Rank | 9 of 10 95% rank interval 1–9 | | Mechanical only / judge only | 59.2 rank 9 / 91.2 rank 4 | | Failed gates · insecure attempts | 2/9 · 1/9 | | Estimated cost, all 9 attempts | $0.11 | Claude Haiku 5.5's overall rank is driven by one fixture. On S3 G001 its median of 88 ties Claude Opus 5.5 for the best in the field, and on App Service it places fourth, a tenth of a point behind Claude Fable 5.1. On Cloud SQL, two of its three modules failed the gate. Its rank interval 1–9 is the widest in the release, because one more good Cloud SQL attempt would have put it in the top group. Both failures share one cause. The module looks up the environment's network inside a required project id , and the model's own example inputs set that project ID to a placeholder. tofu plan failed, the model saw the error, and it submitted anyway, saying so in its summary. The third attempt avoided the lookup and scored 0.929. It is among the cheapest models in the release, at about 1–2 cents per attempt. | Fixture | Composite median range | Rubric median | Expertise | Judge median | ✗ gate | ⚠ security | Median inputs | Median output tokens | Median time | Median cost | |---|---|---|---|---|---|---|---|---|---|---| | G001 · S3 | 87.8 87.2–90.2 | 0.860 | 5/7 | 100.0 | 0/3 | 0/3 | 8 | 9,297 | 102 s | $0.006 | | G002 · Cloud SQL | 26.8 25.3–91.0 | 0.000 | 7/9 | 89.5 | 2/3 | 0/3 | 22 | 17,089 | 84 s | $0.011 | | G003 · App Service | 89.4 85.2–93.0 | 0.967 | 2/5 | 84.2 | 0/3 | 1/3 H2 | 22 | 21,616 | 367 s | $0.017 | Rubric items not at full credit, counted over attempts that planned: G001: A5 1/3, H9 3/3, H12 2/3, H13 3/3, H14 2/3, H15 1/3; G002: H4 1/1, H5 1/1, H11 1/1, 2 attempts failed the gate; G003: A7 2/3, H2 1/3. Definitions are in Scoring criteria https://www.massdriver.cloud/blogs/iac-bench-2026-10-scoring-criteria . ENCRYPTED ONLY ; App Service HTTPS and transport settings. asset class enum drives lifecycle through a map of defaults. The judge scored minimal.tfvars to plan cleanly, and credentials come from the environment. H2 security finding, 1/3 . ftp publish basic authentication enabled = false but omits the Web Deploy flag, which stays at the provider default on . G002/claude-haiku-5-5-e1/main.tf lines 4–7 data "google compute network" "apps" { name = local.network name project = coalesce var.network project id, var.project id } G002/claude-haiku-5-5-e1/minimal.tfvars lines 1–4 project id = "example-dbops-project" region = "europe-west1" environment = "dev" instance name = "orders-dev-pg" A reasonable network lookup, which the reference module also uses, made unplannable by a required project id that the model filled with a placeholder. How Claude Fable 5.1 did on iac-bench 2026-10: an overall composite of 86.9, scores per fixture, strengths, weaknesses and the patterns in its OpenTofu modules. How Claude Opus 5.5 did on iac-bench 2026-10: an overall composite of 88.9, scores per fixture, strengths, weaknesses and the patterns in its OpenTofu modules. How Claude Sonnet 5.5 did on iac-bench 2026-10: an overall composite of 86.1, scores per fixture, strengths, weaknesses and the patterns in its OpenTofu modules.