Use when deciding whether to route a task to OpenAI's GPT-5.5 frontier model versus Claude Opus or Sonnet — picking the model lane for infrastructure scripts, CI pipelines, concrete implementation, analytical code review, security review, or CLI/terminal-heavy work, and weighing GPT-5.5's context window, pricing, and per-benchmark strengths against the Claude tiers for the same task. Covers the decision-useful capability and pricing facts, and the boundary against the Claude routing skills. Do NOT use for running the GPT model through a harness (use `codex`), for choosing among the Claude tiers themselves (use `claude-opus` / `claude-sonnet`), or for routing among local skills at request time (use `skill-router`). Do NOT use for resume my last Codex session and keep going. Do NOT use for is Opus or Sonnet the right Claude tier for this? Do NOT use for which of my skills handles webhook tasks?
What it is: GPT-5.5 is OpenAI's current GPT frontier model generation (~1M context). This skill is the routing decision: when a task's shape favors GPT-5.5's strengths enough to send it there instead of a Claude tier.
Mental model: Treat frontier models as specialists with overlapping but non-identical edges. GPT-5.5 and Claude Opus are within a couple of points on aggregate coding benchmarks, but they diverge by task type: GPT-5.5 leads on terminal/CLI workflows and precise tool-use; Claude Opus leads on architecture and complex multi-file resolution. Routing is matching the task's dominant axis to the model that owns that axis — not crowning one model "best."
Why it exists: "Just use the strongest model" is the wrong default — it ignores that strength is task-shaped, and it ignores cost. An explicit per-task routing rule captures the real per-benchmark splits so concrete-implementation and review work lands on GPT-5.5 while architecture work stays on Opus, instead of being decided by habit.
What it is NOT: It is not the Codex CLI (that is the harness that runs GPT — a separate decision), not the OpenAI SDK/API, and not a claim that GPT-5.5 beats Claude everywhere. It is a model-lane selector, not a leaderboard.
Adjacent concepts: the Codex CLI (the harness for GPT), the Claude tier skills (claude-opus, claude-sonnet, claude-haiku), gemini-pro (the third frontier lane), and skill-router (request-time dispatch among local skills, a different layer).
One-line analogy: Routing GPT-5.5 vs Opus is choosing a surgeon by procedure, not by overall reputation — the more decorated surgeon is still the wrong pick for the operation the other one specializes in.
Common misconception: That the higher SWE-bench Verified score makes GPT-5.5 the universal coding choice. The aggregate margin is within noise and benchmark-contamination caveats; on complex multi-file resolution (SWE-bench Pro) Claude Opus leads, so the verified-bench headline does not generalize to architecture-heavy work.
Coverage
The per-task routing decision: when a task's dominant axis favors GPT-5.5 enough to send it there instead of a Claude tier
The per-benchmark splits that drive routing — GPT-5.5 leading Terminal-Bench / CLI / tool-use and concrete implementation, Claude Opus leading SWE-bench Pro and broad architecture
The decision-useful capability and pricing facts: ~1M context window, the >272K-token surcharge, standard / Pro / cached pricing, and how they compare to the Claude tiers
The subscription-vs-API distinction: why headline per-token USD is notional on a flat subscription and must not drive routing alone
The boundary against adjacent skills: the Codex harness (running GPT), the Claude tier skills (intra-Claude routing), gemini-pro (the third frontier lane), and skill-router (request-time local-skill dispatch)
What this skill is NOT: the Codex CLI, the OpenAI SDK, a leaderboard, or a claim that GPT-5.5 beats Claude everywhere
Philosophy of the skill
Routing is matching, not ranking. The tempting shortcut — "use the strongest model and stop thinking" — is wrong twice over: it treats model strength as a single scalar when it is actually a profile of task-shaped edges, and it ignores cost entirely. This skill refuses both errors. It insists the routing primitive is the task's dominant axis, not a model's aggregate benchmark rank, because two near-equal models split by task type and the per-task winner can flip even when one model leads on paper. It also refuses to let headline price be the primitive: on a flat subscription the per-token USD is notional, and even on API the capability-per-task delta usually dominates the price delta. The discipline is to name the task's dominant axis first, route to the model that owns that axis, and treat both "always strongest" and "always cheapest" as reflexes to override.
When to route to GPT-5.5 (decision)
Send a task to GPT-5.5 when its dominant axis is one GPT-5.5 owns:
Task axis
Route
Why
Infrastructure scripts, CI pipelines
GPT-5.5
Strong on concrete, well-specified implementation
CLI / terminal-heavy iteration (planning, tool coordination in a shell)
GPT-5.5
Leads Terminal-Bench 2.0 (82.7%) — its most decisive win
Analytical code review of a diff; security review
GPT-5.5
Strong analytical diff-reading and threat-surface work
Precise tool use + file navigation
GPT-5.5
Holds a slight edge here
Broad architectural reasoning across a large codebase
Claude Opus
Opus leads architecture work
Complex multi-file GitHub issue resolution
Claude Opus
Opus leads SWE-bench Pro (64.3% vs 58.6%)
Multi-tool agent coordination
Claude
Claude's tool-coordination edge (MCP Atlas)
Cost-sensitive but still-capable implementation
Claude Sonnet
Cheaper than both frontier tiers — see claude-sonnet
Capability and pricing facts (2026-06-08)
Fact
GPT-5.5
For comparison
Context window
~1,050,000 tokens
Claude Opus 4.8 / Sonnet 4.6: 1M
Standard input / output
$5 / $30 per 1M
Opus 4.8: $5 / $25 · Sonnet 4.6: $3 / $15
Pro tier
$30 / $180 per 1M
—
Cached input
$0.50 / 1M
—
Long-context surcharge
> 272K input → 2× in / 1.5× out (full session)
—
SWE-bench Verified
#1 (88.7%)
Opus 4.7: 87.6% (margin within noise + contamination caveats)
SWE-bench Pro
58.6%
Opus 4.7 leads (64.3%)
Terminal-Bench 2.0
82.7% (leads)
—
On subscription plans (not per-token API), the per-token USD above is notional, not the billed cost — do not route on headline price alone when billing is a flat subscription.
Strengths and weaknesses
Strengths
Leads real CLI/terminal workflows and precise tool-use/file-navigation.
Top-tier on SWE-bench Verified and strong on concrete implementation, infra/CI, analytical and security review.
~1M context for large-context coding and research.
Weaknesses / watch-outs
Higher output price per token than Claude Opus ($30 vs $25); the >272K surcharge compounds long-context cost.
Trails Claude Opus on complex multi-file resolution and broad architecture.
SWE-bench Verified scores at this tier carry contamination caveats — do not over-weight the headline number.
This is a model fact set; the harness you run GPT through (Codex) is a separate decision.
Verification
Use this checklist to confirm a GPT-vs-Claude routing decision is correct and current.
The task's dominant axis was named (infra/CI, CLI/terminal, analytical/security review, architecture, multi-file resolution, multi-tool coordination) BEFORE a model was picked — not chosen by habit or headline rank
The route follows the axis → model map, not "always the strongest model" or "always the cheapest"
The current GPT-5.5 generation, context window, and pricing were re-checked against the provider's live docs — these change fast and a stale number can flip the decision
When cost was a factor, billing basis was checked: on a flat subscription the per-token USD is notional, not the billed cost, and the >272K-token surcharge was accounted for on long-context work
The decision did not silently fold in a harness choice (running GPT) — that is a separate decision routed to the Codex harness skill
A benchmark headline was not over-weighted: near-equal margins and contamination caveats were treated as such, and architecture-heavy work was not routed to GPT on the strength of a verified-bench lead alone
Do NOT Use When
Instead of gpt-5-5
Use
Why
Running GPT through its CLI agent (resume sessions, exec)
codex
That is a harness decision, not a model-routing one
Choosing among the Claude tiers (Opus vs Sonnet vs Haiku)
claude-opus / claude-sonnet / claude-haiku
Intra-Claude routing, not GPT-vs-Claude
Considering the Google frontier lane
gemini-pro
A different model family's routing
Dispatching among local skills at request time
skill-router
A different layer entirely
References
references/model-facts.md — verified GPT-5.5 capability, pricing, and benchmark facts (2026-06-08) with sources
Sibling skills claude-opus, claude-sonnet, gemini-pro — the other frontier lanes this routing decision weighs against; codex — the harness that runs GPT
Skill Graph context
Classification
Subject: agent-ops (also: ai-engineering)
Public: true
Domain: agent/models
Scope: Deciding when to route a task to OpenAI's GPT-5.5 frontier tier versus Claude Opus or Sonnet: the ~1M context window and its 272K pricing cliff, the $5/$30 standard rates and Pro tier, and the per-benchmark split that says GPT-5.5 leads SWE-bench Verified and Terminal-Bench (CLI/tool-use, infra, concrete implementation, analytical review) while Claude Opus leads SWE-bench Pro and architectural reasoning across large codebases. Portable model-routing knowledge, not anchored to any project. Excludes running GPT through the Codex harness (codex), choosing among Claude tiers (claude-opus/claude-sonnet), and request-time dispatch among local skills (skill-router).
When to use
should this CI pipeline script go to GPT-5.5 or Opus?
is GPT-5.5 or Claude better for an analytical code review of this diff?
what's GPT-5.5's context window and how does its pricing compare to Opus?
Mental model: Frontier models are specialists with overlapping but non-identical edges. Two near-equal models still diverge by task TYPE, so routing means matching a task's dominant axis to the model that owns that axis — not crowning one model 'best overall'.
Purpose: An explicit per-task model-routing rule exists because 'just use the strongest model' ignores that strength is task-shaped and ignores cost. The rule captures real per-benchmark splits so concrete-implementation, CLI/terminal, and analytical-review work lands on the model that leads those axes while architecture-heavy work stays on the model that leads there.
Boundary: This decides WHICH frontier MODEL gets a task (the GPT line vs the Claude tiers). It is NOT the CLI harness that runs the model, NOT the model's SDK/API, NOT intra-Claude tier selection, and NOT request-time dispatch among local skills.
Analogy: Choosing the model is choosing a surgeon by procedure, not by overall reputation — the more decorated surgeon is still the wrong pick for the operation the other one specializes in.
Common misconception: That a higher headline coding-benchmark score makes one model the universal coding choice. Aggregate margins are often within noise and carry contamination caveats; on complex multi-file resolution the ranking can flip, so a single verified-bench headline does not generalize to architecture-heavy work.
Grounding
Mode: hybrid
Truth sources: skills/agent-ops/gpt-5-5/references/model-facts.md
Keywords
when to use gpt-5.5, gpt-5.5 vs claude opus, route task to gpt, gpt-5.5 for code review, gpt-5.5 infrastructure scripts, gpt-5.5 context window pricing, gpt-5.5 security review, model routing gpt claude, swe-bench gpt vs claude, gpt-5.5 terminal bench
1---2name: gpt-5-53description: Use when deciding whether to route a task to OpenAI's GPT-5.5 frontier model versus Claude Opus or Sonnet — picking the model lane for infrastructure scripts, CI pipelines, concrete implementation, analytical code review, security review, or CLI/terminal-heavy work, and weighing GPT-5.5's context window, pricing, and per-benchmark strengths against the Claude tiers for the same task. Covers the decision-useful capability and pricing facts, and the boundary against the Claude routing skills. Do NOT use for running the GPT model through a harness (use `codex`), for choosing among the Claude tiers themselves (use `claude-opus` / `claude-sonnet`), or for routing among local skills at request time (use `skill-router`). Do NOT use for resume my last Codex session and keep going. Do NOT use for is Opus or Sonnet the right Claude tier for this? Do NOT use for which of my skills handles webhook tasks?4license: MIT5---67# GPT-5.589## Concept of the skill1011**What it is:** GPT-5.5 is OpenAI's current GPT frontier model generation (~1M context). This skill is the routing decision: when a task's shape favors GPT-5.5's strengths enough to send it there instead of a Claude tier.1213**Mental model:** Treat frontier models as specialists with overlapping but non-identical edges. GPT-5.5 and Claude Opus are within a couple of points on aggregate coding benchmarks, but they diverge by task *type*: GPT-5.5 leads on terminal/CLI workflows and precise tool-use; Claude Opus leads on architecture and complex multi-file resolution. Routing is matching the task's dominant axis to the model that owns that axis — not crowning one model "best."1415**Why it exists:** "Just use the strongest model" is the wrong default — it ignores that strength is task-shaped, and it ignores cost. An explicit per-task routing rule captures the real per-benchmark splits so concrete-implementation and review work lands on GPT-5.5 while architecture work stays on Opus, instead of being decided by habit.1617**What it is NOT:** It is not the Codex CLI (that is the *harness* that runs GPT — a separate decision), not the OpenAI SDK/API, and not a claim that GPT-5.5 beats Claude everywhere. It is a model-lane selector, not a leaderboard.1819**Adjacent concepts:** the Codex CLI (the harness for GPT), the Claude tier skills (`claude-opus`, `claude-sonnet`, `claude-haiku`), `gemini-pro` (the third frontier lane), and `skill-router` (request-time dispatch among local skills, a different layer).2021**One-line analogy:** Routing GPT-5.5 vs Opus is choosing a surgeon by procedure, not by overall reputation — the more decorated surgeon is still the wrong pick for the operation the other one specializes in.2223**Common misconception:** That the higher SWE-bench Verified score makes GPT-5.5 the universal coding choice. The aggregate margin is within noise and benchmark-contamination caveats; on complex multi-file resolution (SWE-bench Pro) Claude Opus leads, so the verified-bench headline does not generalize to architecture-heavy work.2425## Coverage2627- The per-task routing decision: when a task's dominant axis favors GPT-5.5 enough to send it there instead of a Claude tier28- The per-benchmark splits that drive routing — GPT-5.5 leading Terminal-Bench / CLI / tool-use and concrete implementation, Claude Opus leading SWE-bench Pro and broad architecture29- The decision-useful capability and pricing facts: ~1M context window, the >272K-token surcharge, standard / Pro / cached pricing, and how they compare to the Claude tiers30- The subscription-vs-API distinction: why headline per-token USD is notional on a flat subscription and must not drive routing alone31- The boundary against adjacent skills: the Codex harness (running GPT), the Claude tier skills (intra-Claude routing), `gemini-pro` (the third frontier lane), and `skill-router` (request-time local-skill dispatch)32- What this skill is NOT: the Codex CLI, the OpenAI SDK, a leaderboard, or a claim that GPT-5.5 beats Claude everywhere3334## Philosophy of the skill3536Routing is matching, not ranking. The tempting shortcut — "use the strongest model and stop thinking" — is wrong twice over: it treats model strength as a single scalar when it is actually a profile of task-shaped edges, and it ignores cost entirely. This skill refuses both errors. It insists the routing primitive is the task's *dominant axis*, not a model's aggregate benchmark rank, because two near-equal models split by task type and the per-task winner can flip even when one model leads on paper. It also refuses to let headline price be the primitive: on a flat subscription the per-token USD is notional, and even on API the capability-per-task delta usually dominates the price delta. The discipline is to name the task's dominant axis first, route to the model that owns that axis, and treat both "always strongest" and "always cheapest" as reflexes to override.3738## When to route to GPT-5.5 (decision)3940Send a task to GPT-5.5 when its **dominant axis** is one GPT-5.5 owns:4142| Task axis | Route | Why |43|---|---|---|44| Infrastructure scripts, CI pipelines | **GPT-5.5** | Strong on concrete, well-specified implementation |45| CLI / terminal-heavy iteration (planning, tool coordination in a shell) | **GPT-5.5** | Leads Terminal-Bench 2.0 (82.7%) — its most decisive win |46| Analytical code review of a diff; security review | **GPT-5.5** | Strong analytical diff-reading and threat-surface work |47| Precise tool use + file navigation | **GPT-5.5** | Holds a slight edge here |48| Broad architectural reasoning across a large codebase | **Claude Opus** | Opus leads architecture work |49| Complex multi-file GitHub issue resolution | **Claude Opus** | Opus leads SWE-bench Pro (64.3% vs 58.6%) |50| Multi-tool agent coordination | **Claude** | Claude's tool-coordination edge (MCP Atlas) |51| Cost-sensitive but still-capable implementation | **Claude Sonnet** | Cheaper than both frontier tiers — see `claude-sonnet` |5253## Capability and pricing facts (2026-06-08)5455| Fact | GPT-5.5 | For comparison |56|---|---|---|57| Context window | ~1,050,000 tokens | Claude Opus 4.8 / Sonnet 4.6: 1M |58| Standard input / output | $5 / $30 per 1M | Opus 4.8: $5 / $25 · Sonnet 4.6: $3 / $15 |59| Pro tier | $30 / $180 per 1M | — |60| Cached input | $0.50 / 1M | — |61| Long-context surcharge | > 272K input → 2× in / 1.5× out (full session) | — |62| SWE-bench Verified | #1 (88.7%) | Opus 4.7: 87.6% (margin within noise + contamination caveats) |63| SWE-bench Pro | 58.6% | Opus 4.7 leads (64.3%) |64| Terminal-Bench 2.0 | 82.7% (leads) | — |6566> On subscription plans (not per-token API), the per-token USD above is notional, not the billed cost — do not route on headline price alone when billing is a flat subscription.6768## Strengths and weaknesses6970**Strengths**71- Leads real CLI/terminal workflows and precise tool-use/file-navigation.72- Top-tier on SWE-bench Verified and strong on concrete implementation, infra/CI, analytical and security review.73- ~1M context for large-context coding and research.7475**Weaknesses / watch-outs**76- Higher output price per token than Claude Opus ($30 vs $25); the >272K surcharge compounds long-context cost.77- Trails Claude Opus on complex multi-file resolution and broad architecture.78- SWE-bench Verified scores at this tier carry contamination caveats — do not over-weight the headline number.79- This is a *model* fact set; the *harness* you run GPT through (Codex) is a separate decision.8081## Verification8283Use this checklist to confirm a GPT-vs-Claude routing decision is correct and current.8485- [ ] The task's **dominant axis** was named (infra/CI, CLI/terminal, analytical/security review, architecture, multi-file resolution, multi-tool coordination) BEFORE a model was picked — not chosen by habit or headline rank86- [ ] The route follows the axis → model map, not "always the strongest model" or "always the cheapest"87- [ ] The current GPT-5.5 generation, context window, and pricing were re-checked against the provider's live docs — these change fast and a stale number can flip the decision88- [ ] When cost was a factor, billing basis was checked: on a flat subscription the per-token USD is notional, not the billed cost, and the >272K-token surcharge was accounted for on long-context work89- [ ] The decision did not silently fold in a harness choice (running GPT) — that is a separate decision routed to the Codex harness skill90- [ ] A benchmark headline was not over-weighted: near-equal margins and contamination caveats were treated as such, and architecture-heavy work was not routed to GPT on the strength of a verified-bench lead alone9192## Do NOT Use When9394| Instead of `gpt-5-5` | Use | Why |95|---|---|---|96| Running GPT through its CLI agent (resume sessions, exec) | `codex` | That is a harness decision, not a model-routing one |97| Choosing among the Claude tiers (Opus vs Sonnet vs Haiku) | `claude-opus` / `claude-sonnet` / `claude-haiku` | Intra-Claude routing, not GPT-vs-Claude |98| Considering the Google frontier lane | `gemini-pro` | A different model family's routing |99| Dispatching among local skills at request time | `skill-router` | A different layer entirely |100101## References102103- `references/model-facts.md` — verified GPT-5.5 capability, pricing, and benchmark facts (2026-06-08) with sources104- Sibling skills `claude-opus`, `claude-sonnet`, `gemini-pro` — the other frontier lanes this routing decision weighs against; `codex` — the harness that runs GPT105106## Skill Graph context107108<!-- skill-graph-context:start (generated — do not edit by hand) -->109110**Classification**111- Subject: `agent-ops` (also: `ai-engineering`)112- Public: `true`113- Domain: `agent/models`114- Scope: Deciding when to route a task to OpenAI's GPT-5.5 frontier tier versus Claude Opus or Sonnet: the ~1M context window and its 272K pricing cliff, the $5/$30 standard rates and Pro tier, and the per-benchmark split that says GPT-5.5 leads SWE-bench Verified and Terminal-Bench (CLI/tool-use, infra, concrete implementation, analytical review) while Claude Opus leads SWE-bench Pro and architectural reasoning across large codebases. Portable model-routing knowledge, not anchored to any project. Excludes running GPT through the Codex harness (codex), choosing among Claude tiers (claude-opus/claude-sonnet), and request-time dispatch among local skills (skill-router).115116**When to use**117- should this CI pipeline script go to GPT-5.5 or Opus?118- is GPT-5.5 or Claude better for an analytical code review of this diff?119- what's GPT-5.5's context window and how does its pricing compare to Opus?120121**Not for**122- resume my last Codex session and keep going123- is Opus or Sonnet the right Claude tier for this?124- which of my skills handles webhook tasks?125- Owned by `codex`126- Owned by `claude-opus`127128**Related skills**129- Verify with: `claude-opus`, `codex`130- Related: `codex`, `claude-opus`, `claude-sonnet`, `gemini-pro`, `claude-code`131132**Concept**133- Mental model: Frontier models are specialists with overlapping but non-identical edges. Two near-equal models still diverge by task TYPE, so routing means matching a task's dominant axis to the model that owns that axis — not crowning one model 'best overall'.134- Purpose: An explicit per-task model-routing rule exists because 'just use the strongest model' ignores that strength is task-shaped and ignores cost. The rule captures real per-benchmark splits so concrete-implementation, CLI/terminal, and analytical-review work lands on the model that leads those axes while architecture-heavy work stays on the model that leads there.135- Boundary: This decides WHICH frontier MODEL gets a task (the GPT line vs the Claude tiers). It is NOT the CLI harness that runs the model, NOT the model's SDK/API, NOT intra-Claude tier selection, and NOT request-time dispatch among local skills.136- Analogy: Choosing the model is choosing a surgeon by procedure, not by overall reputation — the more decorated surgeon is still the wrong pick for the operation the other one specializes in.137- Common misconception: That a higher headline coding-benchmark score makes one model the universal coding choice. Aggregate margins are often within noise and carry contamination caveats; on complex multi-file resolution the ranking can flip, so a single verified-bench headline does not generalize to architecture-heavy work.138139**Grounding**140- Mode: `hybrid`141- Truth sources: `skills/agent-ops/gpt-5-5/references/model-facts.md`142143**Keywords**144- `when to use gpt-5.5`, `gpt-5.5 vs claude opus`, `route task to gpt`, `gpt-5.5 for code review`, `gpt-5.5 infrastructure scripts`, `gpt-5.5 context window pricing`, `gpt-5.5 security review`, `model routing gpt claude`, `swe-bench gpt vs claude`, `gpt-5.5 terminal bench`145146<!-- skill-graph-context:end -->
Run npx skillmds@latest add jacob-balslev/gpt-5-5 in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
Use when deciding whether to route a task to OpenAI's GPT-5.5 frontier model versus Claude Opus or Sonnet — picking the model lane for infrastructure scripts, CI pipelines, concrete implementation, analytical code review, security review, or CLI/terminal-heavy work, and weighing GPT-5.5's context window, pricing, and per-benchmark strengths against the Claude tiers for the same task. Covers the decision-useful capability and pricing facts, and the boundary against the Claude routing skills. Do NOT use for running the GPT model through a harness (use `codex`), for choosing among the Claude tiers themselves (use `claude-opus` / `claude-sonnet`), or for routing among local skills at request time (use `skill-router`). Do NOT use for resume my last Codex session and keep going. Do NOT use for is Opus or Sonnet the right Claude tier for this? Do NOT use for which of my skills handles webhook tasks? It is listed under AI & ML on SkillMD.
This skill has not completed SkillMD's automated safety review yet. Capability flags: reads secrets. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free. This skill is licensed under MIT.
jacob-balslev (@jacob-balslev) published this skill. Their other Agent Skills are listed on their SkillMD profile.
jacob-balslev