{"id":6913,"date":"2026-09-20T18:06:09","date_gmt":"2026-09-20T18:06:09","guid":{"rendered":"https:\/\/areeblog.com\/?p=6913"},"modified":"2026-09-20T18:06:09","modified_gmt":"2026-09-20T18:06:09","slug":"stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context","status":"publish","type":"post","link":"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/","title":{"rendered":"StepFun\u2019s 600B Step 5 Targets AI Coding Agents With 1M-Token Context"},"content":{"rendered":"

\"StepFun\u2019s<\/p>\n

Chinese artificial intelligence company StepFun has introduced Step 5 Preview, a new flagship model designed for software engineering, AI agents<\/a>, professional knowledge work and finance. The company says the model is built to handle long-running tasks rather than only generate individual answers.<\/p>\n

Step 5 Preview uses a sparse Mixture-of-Experts architecture with 600 billion total parameters, while 27 billion parameters are active for each token. It supports a 1 million-token context window and accepts both text and image input.<\/p>\n

StepFun says the model is available through its products and API, while its weights are scheduled for release on October 15, 2026.<\/p>\n

Artificial Analysis records Step 5 Preview as a reasoning model released in September 2026. Its current evaluation gives the model an Artificial Analysis Intelligence Index score of 44, with an output speed of 99.8 tokens per second and a measured time to first token of 2.96 seconds.<\/p>\n

The same evaluation records pricing of $1 per 1 million input tokens and $2.70 per 1 million output tokens, with a 95 percent cache discount. Artificial Analysis calculates a cost of $0.71 per Intelligence Index task and says the model generated 160 million output tokens during the full evaluation, compared with a 92 million-token median for the models in the comparison set.<\/p>\n

The 1 million-token context window gives Step 5 Preview a substantially larger working space than many models, allowing long conversations, source material, codebases and intermediate results to remain available within a single context. Artificial Analysis estimates that 1 million tokens correspond to about 1,500 pages of A4 text at 12-point Arial formatting.<\/p>\n

StepFun is placing software engineering at the centre of the model’s agent capabilities. Its internal StepCodeBench benchmark covers 553 independent code repositories, nine task categories, 20 application domains and 33 programming languages.<\/p>\n

The task categories include feature modification, bug repair, refactoring, documentation generation, performance tuning, code generation, CI\/CD operations, code transformation and environment setup.<\/p>\n

The covered application areas include web development, mobile applications, cloud infrastructure, DevOps and CI\/CD, databases, distributed systems, data engineering, machine learning, AI applications, cybersecurity, networking, operating systems, compilers and tooling, developer tools, scientific computing, financial technology, e-commerce, multimedia, game development, and embedded and Internet of Things development.<\/p>\n

The programming languages listed by StepFun include Python, C++, Go, Java, JavaScript, TypeScript, Rust, C, C#, Bash and Shell, Kotlin, Swift, Dart, Objective-C, HTML, Fortran, PHP, Ruby, SQL, R, Julia, MATLAB, Scala, Groovy, Clojure, Haskell, Erlang, Elixir, Lua, Perl, Solidity, Zig and Assembly.<\/p>\n

Step 5 Preview achieved an average-at-four score of 49.0 percent on StepCodeBench. StepFun says the model performs particularly well in bug repair, feature modification and refactoring, while also reporting a meaningful gap to leading models on the most difficult long-running tasks.<\/p>\n

On DeepSWE v1.1, Step 5 Preview scored 67.7 percent, compared with 67.5 percent for Kimi K3, 66.9 percent for GLM-5.3, 74.1 percent for GPT-6 Astra and 74.0 percent for Claude Opus 5.<\/p>\n

On ProgramBench, Step 5 Preview scored 80.5 percent, compared with 77.8 percent for Kimi K3, 72.0 percent for GLM-5.3, 85.4 percent for GPT-6 Astra and 82.3 percent for Claude Opus 5.<\/p>\n

On Terminal-Bench v4, Step 5 Preview recorded 33.3 percent, compared with 12.6 percent for Kimi K3, 41.9 percent for GLM-5.3, 57.9 percent for GPT-6 Astra and 52.3 percent for Claude Opus 5.<\/p>\n

Its score on Agents’ Last Exam CLI, or ALE-CLI, was 29.5 percent, compared with 27.6 percent for Kimi K3, 28.6 percent for GLM-5.3, 33.3 percent for GPT-6 Astra and 28.6 percent for Claude Opus 5.<\/p>\n

On GDPval-AA v2, Step 5 Preview recorded 1,571, compared with 1,548 for Kimi K3, 1,634 for GLM-5.3, 1,580 for GPT-6 Astra and 1,735 for Claude Opus 5. StepFun says those GDPval-AA v2 figures use Artificial Analysis results as of September 19, 2026.<\/p>\n

Other published results include 93.5 percent on GPQA Diamond, 46.5 percent on Humanity’s Last Exam, 85.0 percent on Terminal-Bench v2.1, 58.9 percent on SciCode, 84.7 percent on CyberGym and 83.3 percent on DRACO.<\/p>\n

In finance, Step 5 Preview scored 66.4 percent on FrontierFinance, compared with 64.1 percent for GLM-5.3, 62.6 percent for Kimi K3, 55.0 percent for GPT-6 Astra and 69.7 percent for Claude Opus 5.<\/p>\n

StepFun says FrontierFinance is an external benchmark covering six investment use cases through 220 expert-crafted questions and 11,543 evaluation criteria. Its own FinStepBench suite covers live financial search, corporate valuation and deep research.<\/p>\n

Step 5 Preview scored 74.5 percent on FinStepBench LiveSearch, 60.6 percent on CorporateValuation and 55.8 percent on DeepResearch.<\/p>\n

The company has also used long-running experiments to demonstrate how the model handles tasks that require repeated execution and feedback.<\/p>\n

In one 24-hour experiment, StepFun gave Step 5 Preview access to a single NVIDIA H100 GPU and asked it to optimize an MLA GPU kernel from scratch. The configuration used a head dimension of 512, a batch size of one, 64 attention heads and a sequence length of 8,192 tokens.<\/p>\n

The model repeatedly modified the kernel, ran it and measured throughput. StepFun says that when an implementation reduced performance, the model discarded it and continued from the best-performing version.<\/p>\n

After roughly 22 hours, Step 5 Preview reached a peak of 508 TFLOPS for forward and backward execution, compared with 493 TFLOPS for Claude Opus 5. Each model received four independent attempts, with its best run reported.<\/p>\n

In a second 24-hour experiment, Step 5 Preview was tasked with improving a Qwen3-30B-A3B base model through automated post-training using an API annotator. StepFun says the model decided how to use the annotator and how to revise the post-training data.<\/p>\n

The resulting model reached 60 percent accuracy on the official AIME24 test, up from 53.3 percent before post-training. StepFun says this matched the result achieved by Claude Opus 5 while using fewer annotator tokens.<\/p>\n

StepFun also tested the model outside software development with Pok\u00e9mon Red. Without Pok\u00e9mon-specific optimization, the company says Step 5 Preview sustained more than 3,000 turns and roughly 6 million interaction tokens. By turn 3,082, it had unlocked Cut, earned three Gym Badges and defeated Lt. Surge. StepFun describes that progress as roughly one-third of the game’s main story.<\/p>\n

The company says the model can also work with programmable hardware when provided with documentation and user authorization. In an ESP32-S3 demonstration lasting more than three hours, Step 5 Preview used cameras, COM ports, screenshots and simulated mouse input while modifying and debugging code in response to real errors.<\/p>\n

Beyond coding, StepFun demonstrated professional work involving industrial engineering, manufacturing engineering, mechanical engineering, process engineering, creative production, live production, video editing and pharmacy.<\/p>\n

In one research demonstration, the model handled a climate study covering 1,000 locations over 25 years and 11 variables. StepFun says it coordinated 950 web fetches in a single agent action and assembled 300,000 monthly records before analysing regional differences in solar seasonality.<\/p>\n

Another demonstration produced a 17-sheet diesel surcharge analysis workbook containing source data, cross-series reconciliation, regional panels, formulas and trend models. StepFun says the workbook contained 11,057 rows and was designed to preserve the relationship between source data and the resulting analysis.<\/p>\n

The model also produced an interactive research report containing written analysis, visualisations, data tables, methodological notes and supporting detail.<\/p>\n

Artificial Analysis provides an independent measurement of Step 5 Preview’s performance and cost. The service places the model at 44 on its Intelligence Index and records it as substantially above the median intelligence score of 24 among comparable models, while also noting that its output is unusually verbose.<\/p>\n

Artificial Analysis says Step 5 Preview’s $1 input price is below the $1.88 median in its comparison group, while its $2.70 output price is below the $10 median. The service calculates a blended price of $0.51 per 1 million tokens using a 7:2:1 cache-hit, input and output ratio.<\/p>\n

StepFun’s launch materials describe Step 5 Preview as a model aimed at shifting the trade-off between intelligence and cost. The sparse architecture is central to that approach because the model has 600 billion total parameters but activates 27 billion per token.<\/p>\n

There are still limits to the public information. StepFun has not publicly disclosed the complete architecture, training-token count, training-compute budget, dataset composition or detailed post-training recipe in the launch material used for these figures.<\/p>\n

The distinction between internal and independent evaluations is also important. StepCodeBench and FinStepBench were built by StepFun, while Artificial Analysis and FrontierFinance provide external evaluation points. StepFun’s 24-hour kernel and automated post-training experiments are company-run demonstrations rather than independently reproduced tests.<\/p>\n

Step 5 Preview is currently proprietary, according to Artificial Analysis, with the model weights not yet publicly available. StepFun says it plans to release the weights on October 15, 2026.<\/p>\n

StepFun’s move follows its earlier Step 3.7 Flash model, which the company describes as a 198-billion-parameter sparse multimodal model with about 11 billion active parameters and a 256,000-token context window. Step 5 Preview increases those headline figures to 600 billion total parameters, 27 billion active parameters and a 1 million-token context window.<\/p>\n

StepFun was founded in 2023 and is based in Shanghai. The company was founded by Jiang Daxin, a former Microsoft Research Asia chief scientist and executive. In January 2026, StepFun completed a financing round exceeding RMB 5 billion, or about $718 million, according to TechNode. The round included Shanghai SDIC Leading Fund, China Life Private Equity Investment, Pudong Venture Capital, Xuhui Capital, Wuxi Liangxi Fund, Xiamen ITG Group and Huaqin Technology, while Tencent, Qiming Venture Partners and 5Y Capital also participated.<\/p>\n","protected":false},"excerpt":{"rendered":"

Chinese artificial intelligence company StepFun has introduced Step 5 Preview, a new flagship model designed for software engineering, AI agents, professional knowledge work and finance. The company says the model is built to handle long-running tasks rather than only generate individual answers. Step 5 Preview uses a sparse Mixture-of-Experts architecture with 600 billion total parameters, […]<\/p>\n","protected":false},"author":2,"featured_media":6914,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"content-type":"","_monsterinsights_skip_tracking":false,"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_feature_clip_id":0,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_post_was_ever_published":false},"categories":[164],"tags":[166],"class_list":["post-6913","post","type-post","status-publish","format-standard","has-post-thumbnail","category-tech-updates","tag-ai"],"share_on_mastodon":{"url":"https:\/\/mastodon.social\/@Areeblog\/117304714888081652","error":""},"yoast_head":"\nStepFun\u2019s 600B Step 5 Targets AI Coding Agents With 1M-Token Context - Aree Blog<\/title>\n<meta name=\"description\" content=\"StepFun\u2019s 600B Step 5 targets AI coding agents with a 1M-token context, sparse architecture and long-running tasks.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"StepFun\u2019s 600B Step 5 Targets AI Coding Agents With 1M-Token Context\" \/>\n<meta property=\"og:description\" content=\"StepFun\u2019s 600B Step 5 targets AI coding agents with a 1M-token context, sparse architecture and long-running tasks.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/\" \/>\n<meta property=\"og:site_name\" content=\"Aree Blog\" \/>\n<meta property=\"article:published_time\" content=\"2026-09-20T18:06:09+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/areeblog.com\/wp-content\/uploads\/2026\/09\/images-62.jpeg\" \/>\n\t<meta property=\"og:image:width\" content=\"670\" \/>\n\t<meta property=\"og:image:height\" content=\"458\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"Daniel Chinonso John\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Daniel Chinonso John\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"8 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/areeblog.com\\\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/areeblog.com\\\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\\\/\"},\"author\":{\"name\":\"Daniel Chinonso John\",\"@id\":\"https:\\\/\\\/areeblog.com\\\/#\\\/schema\\\/person\\\/d972222c55618fb0f4b4c0c11ff52f63\"},\"headline\":\"StepFun\u2019s 600B Step 5 Targets AI Coding Agents With 1M-Token Context\",\"datePublished\":\"2026-09-20T18:06:09+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/areeblog.com\\\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\\\/\"},\"wordCount\":1445,\"commentCount\":0,\"image\":{\"@id\":\"https:\\\/\\\/areeblog.com\\\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/areeblog.com\\\/wp-content\\\/uploads\\\/2026\\\/09\\\/images-62.jpeg\",\"keywords\":[\"AI\"],\"articleSection\":[\"Tech Updates\"],\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/areeblog.com\\\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\\\/#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/areeblog.com\\\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\\\/\",\"url\":\"https:\\\/\\\/areeblog.com\\\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\\\/\",\"name\":\"StepFun\u2019s 600B Step 5 Targets AI Coding Agents With 1M-Token Context - Aree Blog\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/areeblog.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/areeblog.com\\\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/areeblog.com\\\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/areeblog.com\\\/wp-content\\\/uploads\\\/2026\\\/09\\\/images-62.jpeg\",\"datePublished\":\"2026-09-20T18:06:09+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/areeblog.com\\\/#\\\/schema\\\/person\\\/d972222c55618fb0f4b4c0c11ff52f63\"},\"description\":\"StepFun\u2019s 600B Step 5 targets AI coding agents with a 1M-token context, sparse architecture and long-running tasks.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/areeblog.com\\\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/areeblog.com\\\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/areeblog.com\\\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\\\/#primaryimage\",\"url\":\"https:\\\/\\\/areeblog.com\\\/wp-content\\\/uploads\\\/2026\\\/09\\\/images-62.jpeg\",\"contentUrl\":\"https:\\\/\\\/areeblog.com\\\/wp-content\\\/uploads\\\/2026\\\/09\\\/images-62.jpeg\",\"width\":670,\"height\":458,\"caption\":\"StepFun\u2019s 600B Step 5 Targets AI Coding Agents With 1M-Token Context\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/areeblog.com\\\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/areeblog.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"StepFun\u2019s 600B Step 5 Targets AI Coding Agents With 1M-Token Context\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/areeblog.com\\\/#website\",\"url\":\"https:\\\/\\\/areeblog.com\\\/\",\"name\":\"Aree Blog\",\"description\":\"Unfiltered Perspectives, Unstoppable Insights\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/areeblog.com\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/areeblog.com\\\/#\\\/schema\\\/person\\\/d972222c55618fb0f4b4c0c11ff52f63\",\"name\":\"Daniel Chinonso John\",\"description\":\"Daniel Chinonso John is a web designer, penetration tester, and founder of Aree Tech. He writes clear, actionable posts at the intersection of productivity, AI, cybersecurity, and blogging to help readers get things done.\",\"sameAs\":[\"https:\\\/\\\/www.linkedin.com\\\/in\\\/daniel-john-45183a169\\\/\"],\"url\":\"https:\\\/\\\/areeblog.com\\\/author\\\/danojohn55gmail-com\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"StepFun\u2019s 600B Step 5 Targets AI Coding Agents With 1M-Token Context - Aree Blog","description":"StepFun\u2019s 600B Step 5 targets AI coding agents with a 1M-token context, sparse architecture and long-running tasks.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/","og_locale":"en_US","og_type":"article","og_title":"StepFun\u2019s 600B Step 5 Targets AI Coding Agents With 1M-Token Context","og_description":"StepFun\u2019s 600B Step 5 targets AI coding agents with a 1M-token context, sparse architecture and long-running tasks.","og_url":"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/","og_site_name":"Aree Blog","article_published_time":"2026-09-20T18:06:09+00:00","og_image":[{"width":670,"height":458,"url":"https:\/\/areeblog.com\/wp-content\/uploads\/2026\/09\/images-62.jpeg","type":"image\/jpeg"}],"author":"Daniel Chinonso John","twitter_card":"summary_large_image","twitter_misc":{"Written by":"Daniel Chinonso John","Est. reading time":"8 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/#article","isPartOf":{"@id":"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/"},"author":{"name":"Daniel Chinonso John","@id":"https:\/\/areeblog.com\/#\/schema\/person\/d972222c55618fb0f4b4c0c11ff52f63"},"headline":"StepFun\u2019s 600B Step 5 Targets AI Coding Agents With 1M-Token Context","datePublished":"2026-09-20T18:06:09+00:00","mainEntityOfPage":{"@id":"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/"},"wordCount":1445,"commentCount":0,"image":{"@id":"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/#primaryimage"},"thumbnailUrl":"https:\/\/areeblog.com\/wp-content\/uploads\/2026\/09\/images-62.jpeg","keywords":["AI"],"articleSection":["Tech Updates"],"inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/#respond"]}]},{"@type":"WebPage","@id":"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/","url":"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/","name":"StepFun\u2019s 600B Step 5 Targets AI Coding Agents With 1M-Token Context - Aree Blog","isPartOf":{"@id":"https:\/\/areeblog.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/#primaryimage"},"image":{"@id":"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/#primaryimage"},"thumbnailUrl":"https:\/\/areeblog.com\/wp-content\/uploads\/2026\/09\/images-62.jpeg","datePublished":"2026-09-20T18:06:09+00:00","author":{"@id":"https:\/\/areeblog.com\/#\/schema\/person\/d972222c55618fb0f4b4c0c11ff52f63"},"description":"StepFun\u2019s 600B Step 5 targets AI coding agents with a 1M-token context, sparse architecture and long-running tasks.","breadcrumb":{"@id":"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/#primaryimage","url":"https:\/\/areeblog.com\/wp-content\/uploads\/2026\/09\/images-62.jpeg","contentUrl":"https:\/\/areeblog.com\/wp-content\/uploads\/2026\/09\/images-62.jpeg","width":670,"height":458,"caption":"StepFun\u2019s 600B Step 5 Targets AI Coding Agents With 1M-Token Context"},{"@type":"BreadcrumbList","@id":"https:\/\/areeblog.com\/stepfuns-600b-step-5-targets-ai-coding-agents-with-1m-token-context\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/areeblog.com\/"},{"@type":"ListItem","position":2,"name":"StepFun\u2019s 600B Step 5 Targets AI Coding Agents With 1M-Token Context"}]},{"@type":"WebSite","@id":"https:\/\/areeblog.com\/#website","url":"https:\/\/areeblog.com\/","name":"Aree Blog","description":"Unfiltered Perspectives, Unstoppable Insights","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/areeblog.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Person","@id":"https:\/\/areeblog.com\/#\/schema\/person\/d972222c55618fb0f4b4c0c11ff52f63","name":"Daniel Chinonso John","description":"Daniel Chinonso John is a web designer, penetration tester, and founder of Aree Tech. He writes clear, actionable posts at the intersection of productivity, AI, cybersecurity, and blogging to help readers get things done.","sameAs":["https:\/\/www.linkedin.com\/in\/daniel-john-45183a169\/"],"url":"https:\/\/areeblog.com\/author\/danojohn55gmail-com\/"}]}},"jetpack_sharing_enabled":true,"jetpack-related-posts":[{"id":676,"url":"https:\/\/areeblog.com\/deepseek-xiaomi-microsoft-redefine-reasoning-models\/","url_meta":{"origin":6913,"position":0},"title":"DeepSeek, Xiaomi & Microsoft Redefine Reasoning Models","author":"Samuel Ogori","date":"May 2, 2025","format":false,"excerpt":"In the past week, three major players (DeepSeek, Xiaomi, and Microsoft) have each released cutting-edge reasoning models that push the boundaries of what\u2019s possible in math, logic, and code verification. DeepSeek\u2019s gargantuan Prover V2 (671 B parameters) brings formal proof checking to the masses under an MIT license. Xiaomi\u2019s lean\u2026","rel":"","context":"In "Tech Updates"","block_context":{"text":"Tech Updates","link":"https:\/\/areeblog.com\/category\/tech-updates\/"},"img":{"alt_text":"DeepSeek, Xiaomi & Microsoft Redefine Reasoning Models","src":"https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2025\/05\/images.jpeg?resize=350%2C200&ssl=1","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2025\/05\/images.jpeg?resize=350%2C200&ssl=1 1x, https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2025\/05\/images.jpeg?resize=525%2C300&ssl=1 1.5x"},"classes":[]},{"id":4905,"url":"https:\/\/areeblog.com\/z-ai-releases-glm-4-5-series-high-performance-open-source-models-focused-on-agent-capabilities\/","url_meta":{"origin":6913,"position":1},"title":"Z.AI Releases GLM 4.5 Series: High-Performance, Open-Source Models Focused on Agent Capabilities","author":"Samuel Ogori","date":"August 2, 2025","format":false,"excerpt":"Z.AI (formerly Zepoo AI) has launched the GLM 4.5 series, comprising the flagship GLM 4.5 model and the lighter GLM 4.5 Air. Positioned as a significant open-source release in 2025, these models emphasize a balance of performance, efficiency, agent capabilities, and cost. Model Architecture and Efficiency GLM 4.5: A 355\u2026","rel":"","context":"In "Tech Updates"","block_context":{"text":"Tech Updates","link":"https:\/\/areeblog.com\/category\/tech-updates\/"},"img":{"alt_text":"Z.AI Releases GLM 4.5 Series: High-Performance, Open-Source Models Focused on Agent Capabilities","src":"https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2025\/08\/file_00000000f38c6246b5d683f4a05174d4.png?resize=350%2C200&ssl=1","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2025\/08\/file_00000000f38c6246b5d683f4a05174d4.png?resize=350%2C200&ssl=1 1x, https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2025\/08\/file_00000000f38c6246b5d683f4a05174d4.png?resize=525%2C300&ssl=1 1.5x, https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2025\/08\/file_00000000f38c6246b5d683f4a05174d4.png?resize=700%2C400&ssl=1 2x"},"classes":[]},{"id":6598,"url":"https:\/\/areeblog.com\/cloudflare-cut-astros-open-github-issues-by-85-using-a-team-of-ai-agents\/","url_meta":{"origin":6913,"position":2},"title":"Cloudflare Cut Astro’s Open GitHub Issues by 85% Using a Team of AI Agents","author":"Daniel Chinonso John","date":"August 23, 2026","format":false,"excerpt":"Cloudflare and the Astro team have reduced the Astro project's unresolved GitHub issue backlog from more than 200 issues to roughly 30 by using multiple artificial intelligence agents inside GitHub Actions, according to a Cloudflare engineering report. The system was designed around several specialised agents rather than a single autonomous\u2026","rel":"","context":"In "Tech Updates"","block_context":{"text":"Tech Updates","link":"https:\/\/areeblog.com\/category\/tech-updates\/"},"img":{"alt_text":"Cloudflare Cut Astro's Open GitHub Issues by 85% Using a Team of AI Agents","src":"https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/08\/images-36.jpeg?resize=350%2C200&ssl=1","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/08\/images-36.jpeg?resize=350%2C200&ssl=1 1x, https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/08\/images-36.jpeg?resize=525%2C300&ssl=1 1.5x, https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/08\/images-36.jpeg?resize=700%2C400&ssl=1 2x"},"classes":[]},{"id":6471,"url":"https:\/\/areeblog.com\/cloudflare-unveils-kitesurf-a-browser-built-for-ai-agents\/","url_meta":{"origin":6913,"position":3},"title":"Cloudflare Unveils Kitesurf, A Browser Built for AI Agents","author":"Daniel Chinonso John","date":"August 7, 2026","format":false,"excerpt":"Cloudflare has introduced Kitesurf, a cloud-hosted browser designed for autonomous AI agents and built on its Workers platform. The company says the browser is meant for machine-driven web interaction rather than everyday human browsing, and it is being released in beta through Browser Run, Cloudflare\u2019s browser execution service. In its\u2026","rel":"","context":"In "Tech Updates"","block_context":{"text":"Tech Updates","link":"https:\/\/areeblog.com\/category\/tech-updates\/"},"img":{"alt_text":"Cloudflare unveils Kitesurf, a browser built for AI agents","src":"https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/08\/IMG-20260807-WA0011.jpg?resize=350%2C200&ssl=1","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/08\/IMG-20260807-WA0011.jpg?resize=350%2C200&ssl=1 1x, https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/08\/IMG-20260807-WA0011.jpg?resize=525%2C300&ssl=1 1.5x, https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/08\/IMG-20260807-WA0011.jpg?resize=700%2C400&ssl=1 2x, https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/08\/IMG-20260807-WA0011.jpg?resize=1050%2C600&ssl=1 3x"},"classes":[]},{"id":6390,"url":"https:\/\/areeblog.com\/what-nested-ai-agents-mean-for-developers\/","url_meta":{"origin":6913,"position":4},"title":"What Nested AI Agents Mean for Developers","author":"Daniel Chinonso John","date":"July 27, 2026","format":false,"excerpt":"A recent empirical study looked at 1,575 LLM-based agent projects on GitHub and 8,710 related developer discussions, which tells you something important: agent development has moved well beyond toy demos. Developers are now wrestling with real orchestration problems, and nested AI agents are one of the clearest answers emerging from\u2026","rel":"","context":"In "Artificial Intelligence"","block_context":{"text":"Artificial Intelligence","link":"https:\/\/areeblog.com\/category\/artificial-intelligence\/"},"img":{"alt_text":"What Nested AI Agents Mean for Developers","src":"https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/07\/IMG-20260727-WA0001.jpg?resize=350%2C200&ssl=1","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/07\/IMG-20260727-WA0001.jpg?resize=350%2C200&ssl=1 1x, https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/07\/IMG-20260727-WA0001.jpg?resize=525%2C300&ssl=1 1.5x, https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/07\/IMG-20260727-WA0001.jpg?resize=700%2C400&ssl=1 2x, https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/07\/IMG-20260727-WA0001.jpg?resize=1050%2C600&ssl=1 3x"},"classes":[]},{"id":6910,"url":"https:\/\/areeblog.com\/alibabas-new-qwen3-8-omni-flash-pushes-ai-agents-into-long-form-video-and-audio\/","url_meta":{"origin":6913,"position":5},"title":"Alibaba\u2019s New Qwen3.8-Omni-Flash Pushes AI Agents Into Long-Form Video and Audio","author":"Daniel Chinonso John","date":"September 20, 2026","format":false,"excerpt":"Alibaba has introduced Qwen3.8-Omni-Flash, a new native omnimodal AI model designed to move artificial intelligence agents beyond understanding audio and video and toward planning tasks, using tools and completing work. The model accepts text, images, audio and video as input and supports a 1-million-token context window. Alibaba says it is\u2026","rel":"","context":"In "Tech Updates"","block_context":{"text":"Tech Updates","link":"https:\/\/areeblog.com\/category\/tech-updates\/"},"img":{"alt_text":"Alibaba\u2019s New Qwen3.8-Omni-Flash Pushes AI Agents Into Long-Form Video and Audio","src":"https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/09\/alibaba-qwen_GXn5NKdDXY.jpg?resize=350%2C200&ssl=1","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/09\/alibaba-qwen_GXn5NKdDXY.jpg?resize=350%2C200&ssl=1 1x, https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/09\/alibaba-qwen_GXn5NKdDXY.jpg?resize=525%2C300&ssl=1 1.5x, https:\/\/i0.wp.com\/areeblog.com\/wp-content\/uploads\/2026\/09\/alibaba-qwen_GXn5NKdDXY.jpg?resize=700%2C400&ssl=1 2x"},"classes":[]}],"jetpack_featured_media_url":"https:\/\/areeblog.com\/wp-content\/uploads\/2026\/09\/images-62.jpeg","_links":{"self":[{"href":"https:\/\/areeblog.com\/wp-json\/wp\/v2\/posts\/6913","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/areeblog.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/areeblog.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/areeblog.com\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/areeblog.com\/wp-json\/wp\/v2\/comments?post=6913"}],"version-history":[{"count":1,"href":"https:\/\/areeblog.com\/wp-json\/wp\/v2\/posts\/6913\/revisions"}],"predecessor-version":[{"id":6915,"href":"https:\/\/areeblog.com\/wp-json\/wp\/v2\/posts\/6913\/revisions\/6915"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/areeblog.com\/wp-json\/wp\/v2\/media\/6914"}],"wp:attachment":[{"href":"https:\/\/areeblog.com\/wp-json\/wp\/v2\/media?parent=6913"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/areeblog.com\/wp-json\/wp\/v2\/categories?post=6913"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/areeblog.com\/wp-json\/wp\/v2\/tags?post=6913"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}