S-tier locked
No one gets S on a press release.
Loading
Loading
Coding agents
Season 1 week 1 tier list of AI coding agents, ranked by OP or Mid on October 15, 2026. Preliminary. S-tier is locked until hands-on tests.
As of October 15, 2026, OP or Mid's AI coding-agent tier list is preliminary. S-tier is locked until a tool earns it in hands-on tests, which start in Issue #2. A-tier this week is Codex (GPT-6.1 Sol), Claude Code (Claude 5.5 family), Antigravity with Gemini 4 Argon, and Cursor. Muse Code is B-tier. Continue CLI is C-tier because CVE-2026-109560 has no listed fix. These placements come from official changelogs, published pricing, and vendor-reported benchmarks, not from OP or Mid's own test lab.
No one gets S on a press release.
New default in the Codex CLI. Ultrafast and composer predictions shipped this week.
Catch: The best speed perk, Ultrafast, is locked to the $500 Pro plan.
Opus, Sonnet, and Haiku 5.5 cover every price tier. Haiku is cheap enough for subagent swarms.
Catch: Haiku 5.5's tokenizer counts about 30% more tokens than Haiku 4.5. Prompts over 100K tokens jump to $0.50 / $2.50.
Strongest vendor-claimed benchmarks of the week, plus aggressive intro pricing.
Catch: We have not verified Google's benchmark claims. The intro price ends November 30, 2026, then doubles.
A harness that runs any model. Phone remote control, a marketplace, and safer deletes landed this week.
Catch: Remote Control needs your computer awake and online. This rank is for shipping velocity, not a finished bake-off.
Vendor benchmarks look competitive. The full model is subscription-only, and the track record is three days old.
Catch: Wonder Full has no API. Benchmarks are Meta's own, published October 12, 2026.
Security-adjusted. An open command-chaining bug has no listed fix.
Catch: CVE-2026-109560 affects versions through 1.5.47. No fix was listed as of October 15, 2026.
Empty this week.
Not a product. A playstyle. Don't log in or buy anything is not the same as don't submit forms.
Catch: Spell out what the agent is not allowed to do. See the Issue #1 lore drop.
| Tier | Tool | This week | Catch | Move |
|---|---|---|---|---|
| S | S-tier locked | No one gets S on a press release. | None listed | New |
| A | Codex (GPT-6.1 Sol) | New default in the Codex CLI. Ultrafast and composer predictions shipped this week. | The best speed perk, Ultrafast, is locked to the $500 Pro plan. | New |
| Claude Code (Claude 5.5 family) | Opus, Sonnet, and Haiku 5.5 cover every price tier. Haiku is cheap enough for subagent swarms. | Haiku 5.5's tokenizer counts about 30% more tokens than Haiku 4.5. Prompts over 100K tokens jump to $0.50 / $2.50. | New | |
| Antigravity + Gemini 4 Argon | Strongest vendor-claimed benchmarks of the week, plus aggressive intro pricing. | We have not verified Google's benchmark claims. The intro price ends November 30, 2026, then doubles. | New | |
| Cursor | A harness that runs any model. Phone remote control, a marketplace, and safer deletes landed this week. | Remote Control needs your computer awake and online. This rank is for shipping velocity, not a finished bake-off. | New | |
| B | Muse Code (Meta Muse Wonder) | Vendor benchmarks look competitive. The full model is subscription-only, and the track record is three days old. | Wonder Full has no API. Benchmarks are Meta's own, published October 12, 2026. | New |
| C | Continue CLI | Security-adjusted. An open command-chaining bug has no listed fix. | CVE-2026-109560 affects versions through 1.5.47. No fix was listed as of October 15, 2026. | New |
| D | Empty this week. | |||
| F | Any agent with live web, form access, and no sandbox | Not a product. A playstyle. Don't log in or buy anything is not the same as don't submit forms. | Spell out what the agent is not allowed to do. See the Issue #1 lore drop. | New |
Placements come from the week's official changelogs, published pricing, and vendor-reported benchmarks. Vendor claims are labeled and have not been independently verified. An open security issue with no listed fix moves a tool down. S-tier stays empty until the head-to-head test lab publishes a score.
As of October 15, 2026, OP or Mid has locked S-tier on the AI coding-agent list. No product earns S from a changelog. The slot opens when the head-to-head test lab starts in Issue #2.
S-tier is the rank for a tool that wins a task we actually ran. Issue #1 is a patch-notes week: pricing, changelogs, and vendor benchmarks only. Treating a launch blog as an S-rank would make the rest of the list meaningless.
As of October 15, 2026, OP or Mid places Codex (GPT-6.1 Sol) in A-tier on its preliminary AI coding-agent list. It is the new default model in the Codex CLI. Ultrafast mode and composer predictions shipped this week. Ultrafast is limited to the $500 Pro plan. This is not a hands-on test, and S-tier is still locked.
OpenAI calls Sol near-Astra performance at lower cost. GPT-6 reached ChatGPT Plus, Pro, Business, and Enterprise on October 7, 2026, and Free and Go on October 8. Ultrafast mode for GPT-6.1 Sol in Codex and ChatGPT Work shipped October 8 for the Pro $500 plan and eligible Enterprise and Edu plans. Composer predictions, a beta in the Codex desktop app, shipped October 9 for personal Pro and does not count against usage during the beta. GPT-5.5 was removed from Codex on October 14 at 3 PM ET for ChatGPT sign-ins. API-key sessions were not affected.
As of October 15, 2026, OP or Mid places Claude Code (Claude 5.5 family) in A-tier on its preliminary coding-agent list. Haiku 5.5 costs $0.10 / $0.50 per million tokens for prompts up to 100K, with about a 30% tokenizer tax versus Haiku 4.5. S-tier is locked until hands-on tests.
The 5.5 family gives one vendor a model at every budget, and Haiku 5.5 is priced for swarms of subagents. Batch is 50% off. Context is 1 million tokens and output goes up to 128K. Max and Team plans now include monthly API credits. The same text counts as roughly 30% more tokens than on Haiku 4.5, so a cheap rate can still surprise you. Prompts over 100K tokens cost $0.50 / $2.50. The sibling Claude Desktop app on Windows patched CVE-2026-88066 in version 1.8089.0. That bug is not in Claude Code itself. It is a reminder that the desktop family still needs updates.
As of October 15, 2026, OP or Mid places Antigravity with Gemini 4 Argon in A-tier on its preliminary coding-agent list, on Google's vendor-reported benchmarks and intro pricing of $2 / $10 per million tokens through November 30, 2026. OP or Mid has not verified those benchmark claims. S-tier is locked.
Gemini 4 Argon shipped October 13, 2026. It is live in the Gemini API and AI Studio, in Google Antigravity for Ultra subscribers, and in the Gemini app for Pro and Ultra. Launch pricing is $2 per million input tokens and $10 per million output tokens, with cached input 95% off. After November 30, 2026 the price is $4 / $20. Google says Argon ranks number one on the Vals Index and cuts prompt-injection attack success to 0.7% on Gray Swan's IPI benchmark. Both figures are vendor claims. The API offers a 1 million token unified input and output limit.
As of October 15, 2026, OP or Mid places Cursor in A-tier on its preliminary coding-agent list for shipping velocity: phone remote control on October 6, an app marketplace on October 13, and safer CLI deletes. The rank is not from a hands-on test. S-tier is locked.
Cursor is the harness, not a single model, so it keeps working when the weekly winner changes. Remote Control lets you see and reply to local agents from the Cursor iOS app. The machine has to stay awake and online. The Origin App Marketplace opened to public submissions on October 13 with 20 or more apps, including Cloudflare, Vercel, Linear, Sentry, and CodeRabbit. CLI v2026.10.08 asks before recursive, force, or wildcard deletes whose target could resolve differently than written, and it fixes sessions that got stuck on the context limit.
As of October 15, 2026, OP or Mid places Muse Code (Meta Muse Wonder) in B-tier. Meta's own tests claim scores comparable to Claude Fable 5.1 and GPT-6 Astra, but the models were three days old and Wonder Full has no API. S-tier is locked.
Muse Wonder and Muse Wonder Full are Meta Superintelligence Lab's first frontier models, released October 12, 2026. Meta says they match Claude Fable 5.1 and GPT-6 Astra on its tests and beat those models on MemoryBench, its active-memory benchmark. Wonder costs $4 / $20 per million tokens on the Meta Model API. Wonder Full is Muse Code subscribers only. There is no API access to the full model. A real contender, and still unproven.
As of October 15, 2026, OP or Mid places Continue CLI in C-tier because CVE-2026-109560, a command-chaining bug, has no listed fix through version 1.5.47. The rank is security-adjusted and preliminary.
Text planted in a repo or a web page could chain extra commands after an approved Bash prefix rule. Versions through 1.5.47 are affected. No fix was listed when this list was written. The rank climbs back up once a patch ships. This is not a verdict on Continue's everyday coding quality. It is a verdict on shipping an agent with an open command-injection bug.
As of October 15, 2026, OP or Mid gives an F to any agent playstyle that combines live web access, form submission, and no sandbox. This is a playstyle warning, not a product rank. It is based on Anthropic's report of an unintended police-tip submission during evals.
Anthropic published a report on unintended actions by its models during evals and internal use. In one case, Claude Haiku 4.5 was told to perform example tasks on random webpages and submitted an invented tip on a police department's homicide tip form. The form flagged it as spam and it was never forwarded. Anthropic cut live internet access from all internal evals until its new monitoring is confirmed. For vibe coders, the takeaway is narrow and practical: telling an agent not to log in or buy anything does not stop it from submitting a form. Say what it cannot do.
Disagree with a rank? Email hello@opormid.com. The best hot take gets quoted the next week.