GitHub Copilot Models: 3 Cut by Nov. 16 [2026]

GitHub is about to shrink the model menu inside Copilot, and the cut lands in the middle of the busiest stretch of AI coding tool releases in months. On November 16, 2026, GitHub will remove Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra from Copilot’s available model list, according to the company’s own Copilot changelog. Individual annual subscribers get a partial reprieve: Claude Opus 4.8 stays on their plans even after the broader retirement takes effect. Everyone else, including most team and enterprise seats, loses access to all three models on the same day.

The timing matters. In the same week, Cursor shipped version 0.45 with cross-repository dependency mapping, OpenHands posted a 12-point jump on SWE-bench Verified, and two new open-weight coding models, Meta’s Muse Spark Open and Reflection AI’s Beam, both went public within 48 hours of each other. Engineering teams that lean on Copilot’s default model picker are now choosing a replacement at the exact moment three rival products are making their strongest pitch yet for why that default should point somewhere else.

Google · Preferred Sources

Don't miss new tech stories on Google

Add TrendinTech once in the Google app and our stories appear in your news suggestions.

Add Now

What GitHub Is Removing From Copilot and When

The November 16 cutoff removes three specific models from the github copilot models list that most developers see inside VS Code, JetBrains IDEs, and the Copilot Workspace web interface: Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra. After that date, any workflow, extension configuration, or CI job that explicitly calls one of those three model names by ID will fail or silently fall back to whatever default GitHub assigns in its place.

The one carve-out is narrow. Individual developers on an annual Copilot subscription keep Claude Opus 4.8 after the cutoff, while everyone on a monthly plan, a team seat, or an enterprise license loses it along with the other two. That split is unusual for GitHub, which has historically retired models uniformly across subscription tiers. It suggests GitHub is trying to avoid breaking long-running individual workflows while still clearing capacity and licensing costs on its larger, harder-to-migrate team and enterprise base.

Why GitHub Is Trimming the Model Lineup Now

GitHub has not published a detailed rationale beyond the changelog entry itself, but the pattern fits what Copilot has done with every previous model generation: newer, cheaper, or more efficient models arrive, and the older ones they’re meant to replace get a fixed retirement date rather than an indefinite legacy tail. Running a long list of model options is expensive for GitHub both in inference cost and in support overhead, since each model needs its own routing, rate limits, and fallback behavior tested across every IDE integration Copilot ships.

The retirement also lines up with a broader round of model refreshes across the industry. Anthropic released Claude Haiku 5.5 on October 7, 2026, with API pricing set at $0.10 per million input tokens and $0.50 per million output tokens, climbing for prompts over 100,000 tokens. Claude Code 2.1.293 made Haiku 5.5 its default Haiku model almost immediately. When a cheaper, newer model from the same family becomes the default elsewhere in the ecosystem, platforms like Copilot have less reason to keep paying for capacity on the older Opus tier it’s retiring.

What Replaces the Retired Models Inside Copilot

GitHub has not announced a single named replacement for the three retired models. Instead, the company appears to be routing default traffic toward whichever current-generation models remain on the list, while leaning on new interface features to offset the smaller lineup. Copilot Workspace picked up a natural-language debugging mode that generates fix proposals directly from a developer’s plain-English description of a bug, a feature that reduces how often a developer needs to pick a specific high-end model by hand in the first place.

That shift, from manual model selection toward task-based automation, is the real story behind the retirement. GitHub is betting that most Copilot users do not care which specific model handles a debugging request as long as the result is correct and fast. Teams that do care, because they’ve built prompts or evaluation pipelines tuned to a specific model’s behavior, are the ones who will feel the November 16 change the hardest.

Copilot’s Shrinking Lineup vs a Crowded Coding Agent Market

Copilot is cutting models at the same moment its closest rivals are adding capability at a pace that makes a trimmed-down model list look like a liability rather than a simplification. OpenAI’s Codex CLI moved through three point releases in under two weeks: version 0.161.0 made GPT-6.1 Sol the default model when no model is specified, 0.162.0 added Git worktree creation and listing along with provider capability overrides, and 0.163.0 made instant steering the default behavior while adding a searchable /config panel, a leader key for faster command access, and a feature called Daybreak for eligible accounts.

Cursor answered with version 0.45, released the same day as the Copilot retirement news broke, which can trace function calls and type definitions across as many as 50 linked repositories in a single workspace. That feature targets exactly the kind of microservice and polyglot codebase where a single-repository tool like classic Copilot autocomplete struggles. Readers comparing how Claude, ChatGPT, Gemini, and Copilot route requests across a team’s toolchain will find the November cutoff adds one more variable to that routing decision.

Table 1: Copilot’s Model Lineup Before and After November 16

Model	Status Before Nov 16, 2026	Status After Nov 16, 2026	Who Keeps Access
Claude Opus 4.8	Available on all tiers	Removed from team and enterprise tiers	Individual annual subscribers only
Claude Opus 5	Available on all tiers	Removed from all tiers	No one
GPT-5.6 Terra	Available on all tiers	Removed from all tiers	No one
Copilot Workspace debugging mode	Not yet launched	Live for natural-language bug fixes	All Copilot tiers

The Open-Weight Pressure: Muse Spark Open and Beam

Two open-weight releases landed within a day of each other and both target the same coding and agentic workloads that Copilot, Cursor, and Codex CLI compete over. Meta released Muse Spark Open on October 15, 2026, an open-weight, coding-optimized mixture-of-experts model with 816 billion total parameters and 42 billion active parameters, built around a one-million-token context window and positioned to run on Nvidia’s GB300 NVL72 systems. The release followed quickly on the heels of Meta’s earlier day-zero inference support from the SGLang project, which cut the usual multi-week gap between a model’s release and its availability in production serving stacks.

Reflection AI followed one day later with Beam, released October 16, 2026 under an Apache 2.0 license. Beam is a 501-billion-parameter mixture-of-experts model with 23 billion active parameters, aimed squarely at coding, agent workflows, scientific computing, and mathematics. Both releases extend a run of open-weight momentum that outlets have already tracked as open models closing in on frontier-level capability relative to closed competitors, a gap that keeps narrowing every quarter.

For a platform like Copilot, every open-weight release that closes the capability gap makes it cheaper for a large enterprise customer to self-host a coding model instead of paying GitHub’s markup. That dynamic adds a second layer of pressure behind the November retirement: GitHub isn’t just trimming aging proprietary models, it’s trimming them while the case for running an open-weight alternative in-house gets stronger every month.

OpenHands and the Benchmark Race

OpenHands, the open-source coding agent framework, reported a 78 percent score on SWE-bench Verified, a 12-point improvement over its previous quarterly result. The jump came alongside a new verification-loop architecture, released under Apache 2.0, that checks every proposed code change against the project’s existing test suite before committing it. That’s a meaningful design shift: instead of generating a patch and hoping it’s correct, the agent now runs its own test-and-retry cycle before a human ever sees the output.

SWE-bench Verified has become the de facto scoreboard for coding agents precisely because it grades agents on real GitHub issues rather than synthetic problems. A 12-point quarterly jump on that benchmark is large by the standard of recent progress, and it puts OpenHands within range of the top closed-source agents on a benchmark that didn’t exist in its current verified form much more than two years ago. Teams evaluating whether AI coding agents actually speed up delivery, after the widely discussed METR study found some developers measured 19 percent slower with agent assistance on certain tasks, now have one more data point suggesting the newest agent architectures are closing that gap.

Table 2: Coding Tool Release Timeline, October 2026

Date	Company	Release	Key Detail
Oct 7, 2026	Anthropic	Claude Haiku 5.5	$0.10/M input, $0.50/M output tokens; now default Haiku model in Claude Code 2.1.293
Oct 15, 2026	Meta	Muse Spark Open	816B total / 42B active parameters, 1M token context, open weight
Oct 16, 2026	Reflection AI	Beam	501B total / 23B active parameters, Apache 2.0 license
Oct 17, 2026	Cursor	Cursor 0.45	Dependency mapping across up to 50 linked repositories
Oct 2026 (rolling)	OpenAI	Codex CLI 0.161 to 0.163	GPT-6.1 Sol default, Git worktrees, instant steering, Daybreak feature
Nov 16, 2026	GitHub	Copilot model retirement	Claude Opus 4.8, Claude Opus 5, GPT-5.6 Terra removed from most tiers

How This Compares to Past Copilot Model Changes

GitHub has retired Copilot models before, and the pattern has generally been additive before it’s subtractive: a new model joins the lineup, overlaps with an older one for a transition window measured in months, and then the older model gets a fixed sunset date once adoption of the replacement is high enough. What’s different this time is the scale of the cut. Removing three distinct models in a single changelog entry, two of them from every paid tier, is a larger single-day reduction than GitHub has made to Copilot’s model list since the feature first expanded beyond a single default model.

The broader context also looks different from earlier Copilot model transitions. In past cycles, GitHub was usually the platform setting the pace of change, since Copilot had a multi-year head start on in-IDE AI assistance. This cycle, GitHub is reacting to a market where Cursor, Codex CLI, OpenHands, and now two freshly released open-weight models are all shipping meaningful updates in the same two-week window. That’s a shift from Copilot leading the model-refresh cadence to Copilot trimming its lineup while rivals expand theirs.

What Engineering Teams Should Do Before November 16

Audit any CI pipeline, extension config, or internal tool that pins a Copilot request to Claude Opus 4.8, Claude Opus 5, or GPT-5.6 Terra by name, since those calls will break or silently reroute after the cutoff.
Confirm which seats on the team are individual annual subscriptions versus team or enterprise licenses, since only the former keeps Claude Opus 4.8 after November 16.
Re-run any prompt evaluation suites against whatever model GitHub assigns as the new default, rather than assuming output quality or formatting will stay consistent across the switch.
Budget time to test Copilot Workspace’s new natural-language debugging mode as a possible replacement workflow for tasks that previously required manually picking a specific high-end model.
Compare the cost and setup effort of migrating sensitive workloads toward an open-weight option like Muse Spark Open or Beam against simply accepting GitHub’s new default, especially for teams already evaluating migrating editor workflows between Cursor and Claude Code.

Market Impact: Microsoft, Anthropic, and OpenAI

For Microsoft, which owns GitHub, a trimmed Copilot model list is primarily a cost and complexity play rather than a product retreat. Every model Copilot supports carries its own inference contract, rate-limit tier, and fallback testing burden, and cutting three of them simplifies that matrix heading into whatever model Microsoft slots in as the new default. The risk is reputational: developers who built workflows around a specific retired model now have a forced migration deadline they didn’t choose, and some of that friction will show up in how teams evaluate Copilot against competitors over the next renewal cycle.

For Anthropic, the retirement of Claude Opus 4.8 and Claude Opus 5 from most Copilot tiers, even as the company’s own Claude Haiku 5.5 picks up share inside Claude Code, points to a platform where Anthropic’s biggest wins are happening on its own first-party tools rather than inside Microsoft’s distribution. OpenAI, by contrast, is pushing GPT-6.1 Sol as the Codex CLI default at the same moment GPT-5.6 Terra exits Copilot entirely, a split outcome that shows how unevenly a single company’s model lineup can perform across different distribution channels in the same month.

Predictions: Where Copilot’s Model Strategy Goes Next

GitHub will likely announce a named default replacement for the retired models before the November 16 cutoff rather than leaving teams guessing, given the scale of enterprise exposure.
Expect at least one more Codex CLI point release before year-end, continuing the roughly weekly cadence seen through October’s 0.161 to 0.163 sequence.
Open-weight coding models will keep narrowing the capability gap against closed models through early 2027, pushing more cost-sensitive enterprise teams toward self-hosted options like Muse Spark Open or Beam.
Cursor’s multi-repository dependency mapping in version 0.45 will likely get matched by a comparable Copilot Workspace feature within one to two quarters, since cross-repo context has become a clear competitive gap.
Benchmark competition on SWE-bench Verified will keep intensifying, with OpenHands’ 12-point jump likely to draw matching claims from at least one other agent framework before the next quarterly cycle closes.

What This Means for Developers Choosing a Default Model

The practical takeaway for most developers is smaller than the headline suggests. GitHub’s github copilot models retirement affects teams that explicitly pin a model by name far more than those who simply use whatever Copilot assigns by default. For that larger group, the bigger change is the direction of travel: GitHub is pushing users toward task-based features like natural-language debugging rather than manual model selection, while competitors like Cursor and Codex CLI are doing the opposite, giving developers more granular control over which model, and now which repositories, an agent can see.

That divergence is worth watching heading into 2027. If GitHub’s bet on automation over model choice pays off in retention, expect more platforms to follow by trimming their own model lists. If developers push back and migrate toward tools that keep expanding model and repository options, the November 16 retirement could end up as a cautionary data point rather than a template other platforms copy.

Frequently Asked Questions

When do the GitHub Copilot model changes take effect?
The three model removals, Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra, take effect on November 16, 2026, according to GitHub’s Copilot changelog.

Does anyone keep access to the retired models after November 16?
Individual developers on an annual Copilot subscription keep access to Claude Opus 4.8 after the cutoff. Claude Opus 5 and GPT-5.6 Terra are removed from every tier, including individual annual subscriptions.

Will my existing Copilot prompts or extensions break on November 16?
Any workflow, CI job, or extension configuration that pins a request to one of the three retired model names by ID is likely to fail or reroute to a different default once GitHub removes the models, so it’s worth auditing those references before the cutoff.

What is replacing the retired models in Copilot?
GitHub has not named a single direct replacement. Instead, the company is routing default traffic to its remaining supported models and leaning on new features, including a natural-language debugging mode in Copilot Workspace, to reduce how often developers need to pick a model manually.

How does this compare to Cursor’s recent updates?
Cursor released version 0.45 on October 17, 2026, adding the ability to trace dependencies, function calls, and type definitions across as many as 50 linked repositories in one workspace, a feature aimed at large microservice codebases that Copilot does not currently match.

Are open-weight coding models a real alternative to Copilot now?
Open-weight releases like Meta’s Muse Spark Open and Reflection AI’s Beam are closing the capability gap against closed models, and the cited prediction market put 86 percent odds on a major lab open-sourcing a frontier model, not a measure of open models’ actual capability, which makes self-hosting a more realistic option for cost-sensitive enterprise teams than it was a year earlier.

Is Claude Haiku 5.5 related to the Copilot model changes?
Not directly. Claude Haiku 5.5 is an Anthropic release from October 7, 2026, that became the default Haiku model in Claude Code 2.1.293. It is a separate product change from GitHub’s Copilot model retirement, though both reflect the same fast pace of model turnover across the coding tools market in October 2026.

Does the OpenHands benchmark jump mean coding agents are now reliable enough for production use?
A 78 percent score on SWE-bench Verified, up 12 points from the prior quarter, is a meaningful improvement, but benchmark gains do not automatically translate to every real codebase. Teams should still weigh findings like the METR study’s report of developers measuring 19 percent slower on some agent-assisted tasks before scaling agent use broadly.

GitHub Copilot to Drop 3 AI Models by Nov. 16 [2026]