Pimp My IDE / Garage Dispatch
Back to garage
September 23, 2026 | model releases / harnesses / evaluation

A new model is not a drop-in engine.

Claude Opus 5.5 arrived with a new price, a one-million-token context window, adaptive thinking, and fresh coding claims. Claude Code shipped support on the same day. The release notes make the real lesson plain. The model and its harness change together.

THE TAKE: Do not approve a coding-model swap from a benchmark table. Pin the model ID, harness build, system prompt path, effort setting, tool policy, and budget. Then replay work your team already understands.
Run the Model Swap Dyno

The launch number is a vendor claim.

Anthropic says Opus 5.5 costs $4 per million input tokens and $20 per million output tokens. It lists cache reads at $0.20 per million tokens. The company also reports a one-million-token context window, 128,000-token maximum output, and medium default effort.[1][2]

The launch post reports gains on agentic coding and other benchmarks. It also says benchmark margins have become a less reliable guide to real work. That caveat belongs beside the scores, not under the carpet.[1]

A leaderboard can nominate a model. Only a matched run can approve your swap.

The harness has its own release.

Claude Code 2.1.280 made Opus 5.5 the default Opus model. The same release added a control for the maximum MCP description length and changed telemetry around oversized hook output. It also fixed policy checks for writes through symlinked paths and rebuilt damaged cached message lists in one fullscreen failure case.[3]

OpenCode's GPT-6 support tells the same story from another shop. Release 1.18.30 added an Astra-specific system prompt, updated provider SDKs, and added reasoning-effort variants for supported GitLab models.[4]

Those are not cosmetic adapters. A system prompt changes instructions. An SDK changes request and response handling. Effort changes how much work the model spends. Tool descriptions change what the model can see. A model ID without those facts does not identify the system you tested.

Context size is capacity, not permission.

Anthropic's model documentation lists Opus 5.5 with adaptive thinking, medium default effort, a one-million-token context window, and tool use. The same documentation separates API clients from agent runtimes. Claude Code and the Agent SDK add the loop and tool execution above the raw model API.[2]

A larger window can hold more repository state. It can also ingest more stale files, generated output, secrets, or irrelevant history. Keep retrieval rules and tool permissions under review. Do not turn a larger fuel tank into a wider loading dock.

Run a boring swap test.

  1. Pin the exact model identifier and provider route. Record whether an alias can move.
  2. Pin the harness version. Save the system prompt or vendor release note that describes its model-specific path.
  3. Record effort, context cap, tool list, MCP description cap, retry limit, and spend ceiling.
  4. Replay a small set of tasks with known patches and known failure modes.
  5. Compare patch correctness, review time, tool calls, wall time, input, output, and cache cost.
  6. Promote the new setup only if the receipt beats the old one on the measures your work needs.

The exciting model should earn its key on boring evidence.

Interactive makeover / physical adapter review

Model Swap Dyno.

Traditional purpose replaced: pick the newest model name from a menu. Better version: choose the harness, close four evidence couplers, and copy a swap card that keeps the model, adapter, operating policy, and replay result separate.

Choose the adapter

The harness choice changes the receipt. It does not close any evidence coupler.

Coding harness
Evidence couplers
DRAFT2 / 4 couplers
The adapter is named, not proven.

Model identity and harness build are pinned. Operating policy and a matched replay still need evidence.

Why it is better: one native state model drives the wheel, coupler, verdict, and handoff card. The final detent says REVIEW GATE READY. It does not claim that the new model won or that a production route changed.

Sources read, not vibes

Open the source log
  1. Anthropic, "Claude Opus 5.5": launch claims, pricing, cache-read pricing, reported speed and cost changes, benchmark table, benchmark caveat, and availability.
  2. Claude Platform documentation, models overview: current model ID, context and output limits, default effort, thinking mode, pricing, and the distinction between API clients and higher-level agent runtimes.
  3. Claude Code 2.1.280 release notes: Opus 5.5 support, MCP description-length control, hook-output telemetry, symlink policy fix, and session-cache repair.
  4. OpenCode 1.18.30 release notes: Astra-specific system prompt, provider SDK updates, preserved provider model identifiers, and reasoning-effort variants.
  5. Hacker News discussion, September 23, 2026: discovery and public reaction. Comment scores and opinions are not evidence for the technical claims above.

Source boundary: Anthropic's prices and benchmark results are first-party reports. Claude Code and OpenCode release notes verify what those harnesses say they changed. The swap procedure and Model Swap Dyno are Pimp My IDE's review method. They do not certify a model, harness, or provider.