Hi everyone! Step 3.7 Flash is basically an efficiency-first model for real-world agents. It’s fast (up to 400 TPS), handles vision + tool use quite well, and stays relatively stable on longer agent runs. Also plays nicely with common harnesses like @Claude Code , @OpenClaw and @Kilo Code . The weights are also open under Apache 2.0. Love this kind of Flash model: not trying to be the biggest, but trying to be the one agents can actually run a lot.
Human Signal
创始人 / Maker 评论
优先展示 Product Hunt 上对理解条目有帮助的人类文本。
Topics
Open SourceArtificial IntelligenceGitHubDevelopment
痛点分析发布于 2026/05/30
痛点为 AI 基于上游原始证据的初步提炼；未包含额外中国市场检索。
痛点
开发者在使用开源模型构建实际代理（agent）时，面临模型速度慢、视觉与工具使用能力弱、长任务稳定性差的问题。现有模型要么体积过大导致推理延迟高（TPS低），要么在视觉理解和工具调用上表现不佳，使得代理无法高效完成多步骤任务。这导致开发者需要花费大量时间进行模型调优、集成不同组件，甚至被迫在速度和能力之间妥协，增加了开发周期和部署复杂度。
Discussion
精选 Product Hunt 讨论
保留原始讨论语境，用来交叉验证上游条目的真实反馈。
@Stepfun 's Step 3.7 Flash is one of the best open-weight models you can run right now. Perfect timing to play with Kilo's new VS Code extension launched this month .
stepfun is an interesting company to watch because they're operating in a space where the obvious question is always why this over Qwen or Mistral for open weight deployments. the vision plus tool use combination at this size is differentiated but the developer ecosystem and tooling support matters as much as the benchmarks at this point. what does the integration story actually look like, is it compatible with existing inference frameworks or does it need custom setup
源数据· Raw Archive
- source
- Product Hunt
- upstream_source
- producthunt_api
- upstream_item_id
- 1158691
- daily_ranking_item_id
- 76b9cd2c-4826-4592-9944-12e1120d68b8
- rank_date
- 2026-05-30
- rank
- 5
- name
- Step 3.7 Flash
- tagline
- Flash-speed agents model that can see and act
- description
- An Apache 2.0 open-weight Flash model for real-world agents. Step 3.7 Flash combines vision, coding, search, tool use, 256K context, ~11B active params, and up to 400 TPS.
- maker_comment
- <p>Hi everyone!<br><br>Step 3.7 Flash is basically an efficiency-first model for real-world agents.<br><br>It’s fast (up to 400 TPS), handles vision + tool use quite well, and stays relatively stable on longer agent runs. Also plays nicely with common harnesses like <a href="https://www.producthunt.com/products/claude-code" target="_blank" rel="noopener">@Claude Code</a>, <a href="https://www.producthunt.com/products/clawdbot-2" target="_blank" rel="noopener">@OpenClaw</a> and <a href="https://www.producthunt.com/products/kilocode" target="_blank" rel="noopener">@Kilo Code</a>. <br><br>The <a href="https://huggingface.co/collections/stepfun-ai/step-37-flash" target="_blank" rel="nofollow noopener noreferrer">weights</a> are also open under Apache 2.0.<br><br>Love this kind of Flash model: not trying to be the biggest, but trying to be the one agents can actually run a lot.</p>
- votes_count
- 153
- comments_count
- 3
- reviews_count
- 0
- featured_at
- 2026-05-30T07:01:00.000Z
- created_at_on_source
- 2026-05-30T07:01:00.000Z
topics
Open SourceArtificial IntelligenceGitHubDevelopment
media / source-specific data
[
{
"url": "https://ph-files.imgix.net/ba3df7fa-39eb-4ea2-a18f-da48c18b7877.jpeg?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/1bbc18de-875f-4b98-a5f2-a74ec3108aa1.png?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/4dabee3b-962b-4258-97bd-844557b65d0f.png?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/0443b5de-92ba-4724-ad1f-07ae47b76af6.png?auto=format",
"type": "image",
"videoUrl": null
}
]raw_payload
{
"post_id": "1158691",
"fetched_at": "2026-05-30T22:00:02.761Z",
"has_post_detail": true,
"snapshot_version": "producthunt_v1",
"has_maker_comment": true,
"maker_comment_source": "maker_match",
"comments_fetch_status": "ok",
"selected_comment_count": 3
}source_raw_snapshot
{
"id": "b40755bb-96ac-48f8-a5bd-a05c9d57df9f",
"daily_ranking_item_id": "76b9cd2c-4826-4592-9944-12e1120d68b8",
"source": "producthunt_api",
"product_hunt_id": "1158691",
"fetched_at": "2026-05-30T22:00:02.761Z",
"post_raw": {
"id": "1158691",
"url": "https://www.producthunt.com/products/step-3-5-flash?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"name": "Step 3.7 Flash",
"media": [
{
"url": "https://ph-files.imgix.net/ba3df7fa-39eb-4ea2-a18f-da48c18b7877.jpeg?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/1bbc18de-875f-4b98-a5f2-a74ec3108aa1.png?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/4dabee3b-962b-4258-97bd-844557b65d0f.png?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/0443b5de-92ba-4724-ad1f-07ae47b76af6.png?auto=format",
"type": "image",
"videoUrl": null
}
],
"tagline": "Flash-speed agents model that can see and act",
"thumbnail": {
"url": "https://ph-files.imgix.net/aa8a3685-a958-4d2c-93db-5d5787887f36.jpeg?auto=format"
},
"votesCount": 153,
"description": "An Apache 2.0 open-weight Flash model for real-world agents. Step 3.7 Flash combines vision, coding, search, tool use, 256K context, ~11B active params, and up to 400 TPS."
},
"post_detail_raw": {
"id": "1158691",
"slug": "step-3-7-flash",
"user": {
"id": "7991605",
"url": "https://www.producthunt.com/@zaczuo?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"name": "Zac Zuo",
"headline": "Flowtica & AI Frontiers",
"username": "zaczuo",
"websiteUrl": "",
"profileImage": "https://ph-avatars.imgix.net/7991605/b36c3e59-002b-4569-a59f-18a042140146.png?auto=format&crop=faces&fit=crop&h=original&w=original",
"twitterUsername": ""
},
"media": [
{
"url": "https://ph-files.imgix.net/ba3df7fa-39eb-4ea2-a18f-da48c18b7877.jpeg?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/1bbc18de-875f-4b98-a5f2-a74ec3108aa1.png?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/4dabee3b-962b-4258-97bd-844557b65d0f.png?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/0443b5de-92ba-4724-ad1f-07ae47b76af6.png?auto=format",
"type": "image",
"videoUrl": null
}
],
"makers": [],
"topics": {
"edges": [
{
"node": {
"id": "89",
"name": "Open Source",
"slug": "open-source"
}
},
{
"node": {
"id": "268",
"name": "Artificial Intelligence",
"slug": "artificial-intelligence"
}
},
{
"node": {
"id": "272",
"name": "GitHub",
"slug": "github"
}
},
{
"node": {
"id": "589",
"name": "Development",
"slug": "development"
}
}
]
},
"website": "https://www.producthunt.com/r/VAWQFMEA3FMTG3?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"createdAt": "2026-05-30T07:01:00Z",
"dailyRank": 5,
"thumbnail": {
"url": "https://ph-files.imgix.net/aa8a3685-a958-4d2c-93db-5d5787887f36.jpeg?auto=format",
"type": "image",
"videoUrl": null
},
"featuredAt": "2026-05-30T07:01:00Z",
"weeklyRank": 41,
"yearlyRank": null,
"monthlyRank": 390,
"scheduledAt": "2026-05-30T07:01:00Z",
"productLinks": [
{
"url": "https://www.producthunt.com/r/P2BU264OSXRMAD?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"type": "Github"
},
{
"url": "https://www.producthunt.com/r/VAWQFMEA3FMTG3?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"type": "Website"
},
{
"url": "https://www.producthunt.com/r/QYPG2CWWQ4KQCB?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"type": "Hugging Face"
}
],
"reviewsCount": 0,
"commentsCount": 3,
"reviewsRating": 0
},
"maker_comment_raw": {
"id": "5410558",
"url": "https://www.producthunt.com/products/step-3-5-flash?comment=5410558&utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"body": "<p>Hi everyone!<br><br>Step 3.7 Flash is basically an efficiency-first model for real-world agents.<br><br>It’s fast (up to 400 TPS), handles vision + tool use quite well, and stays relatively stable on longer agent runs. Also plays nicely with common harnesses like <a href=\"https://www.producthunt.com/products/claude-code\" target=\"_blank\" rel=\"noopener\">@Claude Code</a>, <a href=\"https://www.producthunt.com/products/clawdbot-2\" target=\"_blank\" rel=\"noopener\">@OpenClaw</a> and <a href=\"https://www.producthunt.com/products/kilocode\" target=\"_blank\" rel=\"noopener\">@Kilo Code</a>. <br><br>The <a href=\"https://huggingface.co/collections/stepfun-ai/step-37-flash\" target=\"_blank\" rel=\"nofollow noopener noreferrer\">weights</a> are also open under Apache 2.0.<br><br>Love this kind of Flash model: not trying to be the biggest, but trying to be the one agents can actually run a lot.</p>",
"user": {
"id": "7991605",
"url": "[REDACTED]",
"name": "[REDACTED]",
"headline": null,
"username": "[REDACTED]",
"profile_url": "[REDACTED]",
"website_url": null,
"profile_image": null,
"twitter_username": null
},
"user_id": "7991605",
"is_maker": true,
"parent_id": null,
"created_at": "2026-05-29T12:47:02Z",
"replies_raw": [],
"votes_count": 6
},
"selected_comments_raw": [
{
"id": "5410558",
"url": "https://www.producthunt.com/products/step-3-5-flash?comment=5410558&utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"body": "<p>Hi everyone!<br><br>Step 3.7 Flash is basically an efficiency-first model for real-world agents.<br><br>It’s fast (up to 400 TPS), handles vision + tool use quite well, and stays relatively stable on longer agent runs. Also plays nicely with common harnesses like <a href=\"https://www.producthunt.com/products/claude-code\" target=\"_blank\" rel=\"noopener\">@Claude Code</a>, <a href=\"https://www.producthunt.com/products/clawdbot-2\" target=\"_blank\" rel=\"noopener\">@OpenClaw</a> and <a href=\"https://www.producthunt.com/products/kilocode\" target=\"_blank\" rel=\"noopener\">@Kilo Code</a>. <br><br>The <a href=\"https://huggingface.co/collections/stepfun-ai/step-37-flash\" target=\"_blank\" rel=\"nofollow noopener noreferrer\">weights</a> are also open under Apache 2.0.<br><br>Love this kind of Flash model: not trying to be the biggest, but trying to be the one agents can actually run a lot.</p>",
"user": {
"id": "7991605",
"url": "[REDACTED]",
"name": "[REDACTED]",
"headline": null,
"username": "[REDACTED]",
"profile_url": "[REDACTED]",
"website_url": null,
"profile_image": null,
"twitter_username": null
},
"user_id": "7991605",
"is_maker": true,
"parent_id": null,
"created_at": "2026-05-29T12:47:02Z",
"replies_raw": [],
"votes_count": 6,
"selection_score": 400,
"selection_reason": "maker_comment"
},
{
"id": "5412310",
"url": "https://www.producthunt.com/products/step-3-5-flash?comment=5412310&utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"body": "<p><a href=\"https://www.producthunt.com/products/step-3-5-flash\" target=\"_blank\" rel=\"noopener\">@Stepfun</a>'s Step 3.7 Flash is one of the best open-weight models you can run right now.</p><p></p><p>Perfect timing to play with Kilo's new VS Code extension launched <a href=\"https://www.producthunt.com/products/kilocode/launches/kilo-code-v7-for-vs-code\" target=\"_blank\" rel=\"noopener\">this month</a>.</p>",
"user": {
"id": "25713",
"url": "[REDACTED]",
"name": "[REDACTED]",
"headline": null,
"username": "[REDACTED]",
"profile_url": "[REDACTED]",
"website_url": null,
"profile_image": null,
"twitter_username": null
},
"user_id": "25713",
"is_maker": false,
"parent_id": null,
"created_at": "2026-05-30T13:45:43Z",
"replies_raw": [],
"votes_count": 2,
"selection_score": 400,
"selection_reason": "body_signal_score"
},
{
"id": "5411956",
"url": "https://www.producthunt.com/products/step-3-5-flash?comment=5411956&utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"body": "<p>stepfun is an interesting company to watch because they're operating in a space where the obvious question is always why this over Qwen or Mistral for open weight deployments. the vision plus tool use combination at this size is differentiated but the developer ecosystem and tooling support matters as much as the benchmarks at this point. what does the integration story actually look like, is it compatible with existing inference frameworks or does it need custom setup</p>",
"user": {
"id": "9814309",
"url": "[REDACTED]",
"name": "[REDACTED]",
"headline": null,
"username": "[REDACTED]",
"profile_url": "[REDACTED]",
"website_url": null,
"profile_image": null,
"twitter_username": null
},
"user_id": "9814309",
"is_maker": false,
"parent_id": null,
"created_at": "2026-05-30T10:24:36Z",
"replies_raw": [],
"votes_count": 0,
"selection_score": 400,
"selection_reason": "body_signal_score"
}
],
"topics_raw": [
{
"id": "89",
"name": "Open Source",
"slug": "open-source"
},
{
"id": "268",
"name": "Artificial Intelligence",
"slug": "artificial-intelligence"
},
{
"id": "272",
"name": "GitHub",
"slug": "github"
},
{
"id": "589",
"name": "Development",
"slug": "development"
}
],
"makers_raw": [],
"launch_metadata_raw": {
"slug": "step-3-7-flash",
"daily_rank": 5,
"featured_at": "2026-05-30T07:01:00Z",
"website_url": "https://www.producthunt.com/r/VAWQFMEA3FMTG3?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"weekly_rank": 41,
"yearly_rank": null,
"monthly_rank": 390,
"scheduled_at": "2026-05-30T07:01:00Z",
"reviews_rating": 0,
"created_at_on_product_hunt": "2026-05-30T07:01:00Z"
},
"stats_raw": {
"votes_count": 153,
"reviews_count": 0,
"comments_count": 3,
"reviews_rating": 0
},
"aux_raw": {
"media_extended": [
{
"url": "https://ph-files.imgix.net/ba3df7fa-39eb-4ea2-a18f-da48c18b7877.jpeg?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/1bbc18de-875f-4b98-a5f2-a74ec3108aa1.png?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/4dabee3b-962b-4258-97bd-844557b65d0f.png?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/0443b5de-92ba-4724-ad1f-07ae47b76af6.png?auto=format",
"type": "image",
"videoUrl": null
}
],
"submitter_user": {
"id": "7991605",
"url": "https://www.producthunt.com/@zaczuo?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"name": "Zac Zuo",
"headline": "Flowtica & AI Frontiers",
"username": "zaczuo",
"profileUrl": "https://www.producthunt.com/@zaczuo?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"websiteUrl": "",
"profileImage": "https://ph-avatars.imgix.net/7991605/b36c3e59-002b-4569-a59f-18a042140146.png?auto=format&crop=faces&fit=crop&h=original&w=original",
"twitterUsername": ""
},
"presentation_only": {
"thumbnail_url": "https://ph-files.imgix.net/aa8a3685-a958-4d2c-93db-5d5787887f36.jpeg?auto=format",
"product_hunt_url": "https://www.producthunt.com/products/step-3-5-flash?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"product_hunt_slug": "step-3-7-flash"
},
"unused_upstream_fields": {
"post_aux": {},
"comments_aux": {},
"post_detail_aux": {
"media_raw": [
{
"url": "https://ph-files.imgix.net/ba3df7fa-39eb-4ea2-a18f-da48c18b7877.jpeg?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/1bbc18de-875f-4b98-a5f2-a74ec3108aa1.png?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/4dabee3b-962b-4258-97bd-844557b65d0f.png?auto=format",
"type": "image",
"videoUrl": null
},
{
"url": "https://ph-files.imgix.net/0443b5de-92ba-4724-ad1f-07ae47b76af6.png?auto=format",
"type": "image",
"videoUrl": null
}
],
"thumbnail_raw": {
"url": "https://ph-files.imgix.net/aa8a3685-a958-4d2c-93db-5d5787887f36.jpeg?auto=format",
"type": "image",
"videoUrl": null
},
"presentation_only": {
"product_hunt_slug": "step-3-7-flash"
},
"product_links_raw": [
{
"url": "https://www.producthunt.com/r/P2BU264OSXRMAD?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"type": "Github"
},
{
"url": "https://www.producthunt.com/r/VAWQFMEA3FMTG3?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"type": "Website"
},
{
"url": "https://www.producthunt.com/r/QYPG2CWWQ4KQCB?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"type": "Hugging Face"
}
],
"submitter_user_raw": {
"id": "7991605",
"url": "https://www.producthunt.com/@zaczuo?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+jingx+%28ID%3A+278031%29",
"name": "Zac Zuo",
"headline": "Flowtica & AI Frontiers",
"username": "zaczuo",
"websiteUrl": "",
"profileImage": "https://ph-avatars.imgix.net/7991605/b36c3e59-002b-4569-a59f-18a042140146.png?auto=format&crop=faces&fit=crop&h=original&w=original",
"twitterUsername": ""
},
"unused_upstream_fields": {}
}
}
},
"selection_meta": {
"scoring_basis": "maker first, then non-maker body signal score with original order tie-break",
"first_comment_id": "5410558",
"comments_strategy": "maker_plus_top_discussion",
"detail_fetch_status": "ok",
"top_candidate_count": 3,
"maker_comment_source": "maker_match",
"comments_fetch_status": "ok",
"dropped_comment_count": 0,
"comments_fetched_limit": 25,
"newest_candidate_count": 3,
"selected_comment_count": 3,
"candidate_comment_count": 3,
"selected_comment_reasons": [
{
"id": "5410558",
"score": 400,
"reason": "maker_comment"
},
{
"id": "5412310",
"score": 400,
"reason": "body_signal_score"
},
{
"id": "5411956",
"score": 400,
"reason": "body_signal_score"
}
]
},
"created_at": "2026-05-30T22:00:08.548Z",
"updated_at": "2026-05-30T22:00:08.548Z"
}