--- title: GPT-5.5 'Spud' release announcement | daily.dev description: OpenAI has released GPT-5.5 (codenamed 'Spud') and GPT-5.5 Pro, the first fully retrained base model since GPT-4.5. The model targets agentic and professional... canonical: https://daily.dev/posts/gpt-5-5-spud-release-announcement-tejjeldcr twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: GPT-5.5 'Spud' release announcement | daily.dev og:description: OpenAI has released GPT-5.5 (codenamed 'Spud') and GPT-5.5 Pro, the first fully retrained base model since GPT-4.5. The model targets agentic and professional... og:url: https://daily.dev/posts/gpt-5-5-spud-release-announcement-tejjeldcr og:image: https://api.daily.dev/og/posts/tEjJeLdCr.png og:image:alt: GPT-5.5 'Spud' release announcement og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # GPT-5.5 'Spud' release announcement **[Collections](https://daily.dev/sources/collections)** · 4 min read · 4 upvotes · 2 comments ## Summary OpenAI has released GPT-5.5 (codenamed 'Spud') and GPT-5.5 Pro, the first fully retrained base model since GPT-4.5. The model targets agentic and professional workloads, posting strong benchmark numbers including 82.7% on Terminal-Bench 2.0 and 73.1% on Expert-SWE. It outperforms Claude Opus 4.7 on terminal coding, math, cybersecurity, and long-context retrieval, though Opus 4.7 still leads on SWE-Bench Pro and several other benchmarks. Developers note faster initial response times and better context inference, but some criticize the model for doing the bare minimum rather than anticipating needs. Pricing is $5/M input and $30/M output tokens — roughly double GPT-5.4 — though token efficiency gains often make per-task costs comparable. OpenAI also announced Workspace Agents, a feature for organizations to build, share, and centrally manage AI agents across teams, connecting to Slack, Salesforce, and Gmail. ## Content OpenAI has released GPT-5.5 (internally codenamed "Spud") and GPT-5.5 Pro, the first fully retrained base model since GPT-4.5. Both are available now in ChatGPT, Codex, and the API, with API access having been delayed briefly for additional safety review before going live. ## What it is GPT-5.5 targets agentic and professional workloads: long-running coding tasks, computer use, document processing, and knowledge work. GPT-5.5 Pro is the higher-end variant aimed at multi-step, quality-sensitive work in legal, business, and technical research. Benchmark numbers worth noting: - 82.7% on Terminal-Bench 2.0 - 73.1% on Expert-SWE (58.6% on SWE-Bench Pro) - 84.9% on GDPval - 78.7% on OSWorld-Verified - Topped the Artificial Analysis Intelligence Index by 3 points, breaking a previous three-way tie with Anthropic and Google On Notion's knowledge work benchmark, GPT-5.5 is 33% faster than its predecessor and uses roughly half the tokens on the same tasks. Perplexity reported 56% fewer tokens on complex computer-use workflows. Security firm Xbow found it cut vulnerability miss rates from 40% (GPT-5) to 10%, though their "Mythos-like hacking" comparison is a stretch since Anthropic's Mythos isn't publicly available. Cursor added it as their top CursorBench model at 72.8%. Augment Code reports it tops their internal benchmarks. It's also available through Databricks (governed via Unity AI Gateway), Vercel AI Gateway, and Kilo Gateway. ## Where it wins and where it doesn't GPT-5.5 is stronger than Claude Opus 4.7 on terminal coding, math, cybersecurity, very long-context retrieval, and ARC-style abstract reasoning. Opus 4.7 still leads on SWE-Bench Pro, FinanceAgent, MCP Atlas, GPQA, and Humanity's Last Exam. One thing developers have noticed: GPT-5.5 starts responding faster, without the long pause before output that older OpenAI models had. That latency improvement sounds minor but apparently matters a lot for how the model feels to use day-to-day. The model also does better at inferring context from conversation history, so it can tailor responses without requiring detailed prompting every time. That said, not everyone is happy with it. Some reviewers find it "lazy" — it technically honors the prompt but only barely, doing the minimum rather than anticipating what you actually need. It also struggles to recover when given incorrect context early in a long thread. The practical advice from people who've used it heavily: be specific upfront, start new threads often, and pre-research context before beginning a task. ## Pricing GPT-5.5 costs $5/M input tokens and $30/M output tokens — roughly double GPT-5.4. That's a real price increase, but the token efficiency gains mean the actual cost per completed task is often comparable or lower for hard tasks. On the Artificial Analysis Index, GPT-5.5 at xhigh settings still came out cheaper than Claude Sonnet. GPT-5.5 at medium settings reportedly performs close to GPT-5.4-xhigh at something closer to a mini-model price point. Note: in-memory prompt caching is not supported. ## The bigger picture OpenAI framed this launch partly as a response to Anthropic's growing enterprise traction — Claude's ARR reportedly jumped from $9B to $30B. The company is building toward a unified desktop app merging ChatGPT, Codex, and the Atlas browser agent. Some observers argue the more significant announcement from OpenAI this week wasn't GPT-5.5 at all, but Workspace Agents: a feature that lets organizations build AI agents once, share them across teams, and manage access centrally. It's in research preview for ChatGPT Business accounts and connects to Slack, Salesforce, Gmail, and other tools. The problem it solves is real — most large organizations have multiple teams independently building overlapping AI workflows with no shared management layer. Ethan Mollick's take is probably the most grounded: GPT-5.5 represents genuine progress, but the "jagged frontier" of AI capability persists. Gains are uneven across task types, and no single model dominates everything. GPT-5.4 remains a reasonable cheaper alternative for workloads where the efficiency gains don't offset the price difference. ## Community discussion Top comments from developers on daily.dev. **@fabianletsch** · 1 upvotes > I recommend trying out Kimi K2.6 on openrouter. > > > Its way cheaper and really good. It can be pretty slow though. **@ronydkid** · 1 upvotes > It does have amazing numbers in those benchmarks, but ehh didn't really have a positive experience with it. > I did give in to the hype and resubed to GPT to try it out. I have this big enterprise app that I'm working on, and I gave gpt 5.5 very clear instructions on what it needs to do and how. It failed... miserably. Okay, maybe I'm overreacting, not "miserably," but the output it produced is not close to opus 4.7. Bear in mind, I have a barebones Claude Code setup (no skills/mcps, except for figma which I also gave to codex) so it doesn't skew my evaluation. Gpt 5.5 really failed to adhere... ## Similar posts on daily.dev - [OpenAI launches GPT-5.5, calling it “a new class of intelligence”](https://daily.dev/posts/openai-launches-gpt-5-5-calling-it-a-new-class-of-intelligence--tqwziccn1) · The New Stack · 1 upvotes · 1 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#openai](https://daily.dev/tags/openai), [#gpt](https://daily.dev/tags/gpt) [View this post on daily.dev](https://daily.dev/posts/gpt-5-5-spud-release-announcement-tejjeldcr) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"GPT-5.5 'Spud' release announcement","url":"https://daily.dev/posts/gpt-5-5-spud-release-announcement-tejjeldcr","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/gpt-5-5-spud-release-announcement-tejjeldcr"},"datePublished":"2026-04-23T18:29:36.721Z","dateModified":"2026-04-25T21:14:44.847Z","description":"OpenAI has released GPT-5.5 (codenamed 'Spud') and GPT-5.5 Pro, the first fully retrained base model since GPT-4.5. The model targets agentic and professional...","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":2,"discussionUrl":"https://daily.dev/posts/gpt-5-5-spud-release-announcement-tejjeldcr","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":4},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":2}],"keywords":"llm,openai,gpt","timeRequired":"PT4M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"GPT-5.5 'Spud' release announcement"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/gpt-5-5-spud-release-announcement-tejjeldcr","comment":[{"@type":"Comment","text":"I recommend trying out Kimi K2.6 on openrouter.\nIts way cheaper and really good. It can be pretty slow though.","datePublished":"2026-04-24T13:27:52.072Z","url":"https://daily.dev/posts/tEjJeLdCr#c-sNTQ8bURi","author":{"@type":"Person","name":"Fabian Letsch","url":"https://daily.dev/fabianletsch","image":"https://lh3.googleusercontent.com/a/ACg8ocKR6BVy_wn23EoOKq7-BlszlcXcLmASlnb7l-GtS-q1bePnkaJf=s96-c"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1}},{"@type":"Comment","text":"It does have amazing numbers in those benchmarks, but ehh didn’t really have a positive experience with it.\nI did give in to the hype and resubed to GPT to try it out. I have this big enterprise app that I’m working on, and I gave gpt 5.5 very clear instructions on what it needs to do and how. It failed… miserably. Okay, maybe I’m overreacting, not “miserably,” but the output it produced is not close to opus 4.7. Bear in mind, I have a barebones Claude Code setup (no skills/mcps, except for figma which I also gave to codex) so it doesn’t skew my evaluation. Gpt 5.5 really failed to adhere to instructions and reuse the existing code and components; it even failed to follow a very common pattern across the whole app—use i18n and don’t hardcode literal strings.\nI did end up giving it a vague description of another app I’m trying to build (personal) and asked it to create a landing page. It did surprise me with the output there, as it was much better than what opus (both in claude code and claude design) managed to produce. However, I noticed that an identical design was shared in the announcement of gpt 5.5.","datePublished":"2026-04-25T00:25:32.937Z","url":"https://daily.dev/posts/tEjJeLdCr#c-RLJCtHYLX","author":{"@type":"Person","name":"ronydkid","url":"https://daily.dev/ronydkid","image":"https://media.daily.dev/image/upload/s--x9KBu8mL--/f_auto/v1762323712/avatars/avatar_OjtjX8oxAYphkqEyy5sii?_a=BAMAK+ZW0"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1}}]} ```