Claude Haiku 5.5 reaches GitHub Copilot with a 100K-token price threshold

The release is broad, but rollout and policy still decide who sees it

GitHub made Claude Haiku 5.5 generally available in GitHub Copilot on October 7. The listed plans are Copilot Pro, Pro+, Max, Business and Enterprise. GitHub lists the model in VS Code, Visual Studio, Copilot CLI, the cloud agent, the Copilot app, github.com, GitHub Mobile, JetBrains IDEs, Xcode and Eclipse.

That client list does not mean every eligible user sees the model immediately. GitHub says rollout is gradual, and its supported‑model reference says availability varies by plan and client. Business and Enterprise administrators can also disable the model through Copilot model policy. A missing picker entry can therefore mean rollout delay, client‑version drift or policy, and those causes need different fixes.

For deployment, test one eligible individual account and one managed account. Record the plan, client and version, then confirm whether the model appears and whether policy permits it. Avoid changing enterprise policy simply because one client has not received a gradual rollout.

GitHub positions Haiku for fast work, but the comparison is not an independent benchmark

GitHub describes Claude Haiku 5.5 as a lightweight model for subagents, quick edits and terminal tasks. It also says early testing matched Claude Sonnet 5 on many coding tasks while using fewer tokens and steps. The announcement does not publish the task set, scoring method, sample size or raw results, so teams should treat that comparison as a vendor claim rather than a purchasing benchmark.

A useful evaluation uses the work you actually route to a fast model: a small code edit, a terminal diagnosis, a bounded repository search and a delegated subtask. Keep the prompt, repository state and acceptance checks fixed. Compare completion quality, elapsed time, input/output tokens and retries with the model you currently use. A cheaper rate is not a saving if failed attempts or human repair erase the difference.

The release also supports configurable reasoning and extended context in selected clients according to GitHub's model table. Those options change consumption, so a comparison should record them rather than attributing every difference to the model name.

The 100K threshold changes all four published token rates

GitHub's current Copilot model‑pricing table lists Claude Haiku 5.5 at $0.10 per million input tokens, $0.01 per million cached input tokens, $0.125 per million cache‑write tokens and $0.50 per million output tokens in the default tier at or below 100,000 input tokens. Above that threshold, the listed long‑context rates are $0.50, $0.05, $0.625 and $2.50 respectively. Each published rate is five times the default‑tier rate.

The threshold is an input‑token threshold, not a monthly allowance. GitHub converts token cost into AI Credits at $0.01 per credit, and plan allowances and overage rules still apply. A team forecasting cost should separate input, cached input, cache writes and output rather than multiply a request count by one headline price.

For a simple uncached illustration, 80,000 input tokens and 10,000 output tokens at the default rates total about $0.013. At the long‑context rates, 120,000 input tokens and 10,000 output tokens total about $0.085. Those examples omit cached input and cache writes and are calculations from the published table, not bills or predictions.

Default enablement can make a new model available without a separate rollout ticket

For Copilot Business and Enterprise, GitHub says new models are automatically enabled when default model enablement is on, unless an administrator has disabled the global default or the specific model. That behavior can surprise teams that expect every model addition to require an explicit enable action.

Before broad use, check the organization or enterprise model policy and record whether default enablement is active. If the model is allowed, confirm which clients expose it and whether existing usage controls apply. If the model is disabled, document the reason and owner rather than asking developers to work around policy through another client.

Model access is only one governance layer. Repository permissions, agent tool permissions, network controls and review requirements continue to apply. Follow GitHub's OpenTools organization page for related product updates, but verify current GitHub policy before changing a managed deployment.

A small acceptance matrix prevents availability and cost assumptions

Use a four‑column acceptance record: account plan, client and version, effective model policy, and observed picker availability. Add the selected context tier and reasoning setting for any cost test. This separates release eligibility from actual delivery and keeps a long‑context request from being compared with a default‑tier run.

For production work, start with low‑risk fast tasks and retain the prior model as a fallback until the new cohort passes quality checks. Review at least one failed or escalated task, not only successful completions. If the model is absent, first compare policy and supported‑client evidence; if it is present but performance is weak, examine task fit and settings before drawing a general conclusion.

Figure: The three gates between announcement and an accepted Copilot rollout. Sources: GitHub's launch post, Copilot model pricing and supported models. Figure by OpenTools Team; no third‑party expressive material used.

Share this article

Related News

GitHub puts shared Copilot Automations under repository ownership

Microsoft’s Copilot harness comes to Foundry for managed agent execution

GitHub Copilot adds server authentication with a billable read-only permission

Claude Haiku 5.5 reaches GitHub Copilot with a 100K-token price threshold

The release is broad, but rollout and policy still decide who sees it

GitHub positions Haiku for fast work, but the comparison is not an independent benchmark

The 100K threshold changes all four published token rates

Default enablement can make a new model available without a separate rollout ticket

A small acceptance matrix prevents availability and cost assumptions

Sources

Tags

Claude Haiku 5.5 reaches GitHub Copilot with a 100K-token price threshold

The release is broad, but rollout and policy still decide who sees it

GitHub positions Haiku for fast work, but the comparison is not an independent benchmark

The 100K threshold changes all four published token rates

Default enablement can make a new model available without a separate rollout ticket

A small acceptance matrix prevents availability and cost assumptions

Sources

Tags