Bolt.new, the browser-based AI app builder from StackBlitz (led by co-founder and CEO Eric Simons), is rolling out Bolt Forge, a new agent that sits in the model picker alongside Standard and Max and runs on open-weight models instead of the closed frontier systems that power the rest of the product. Individual Pro subscribers get up to 50x their normal usage allowance on Forge at no extra cost through October 14, and when the allowance runs out the account rolls back to Standard rather than incurring overage charges.
Under the hood, Forge routes work across GLM 5.3 Flash as the default, with GLM 5.3, Kimi K3, and DeepSeek v4 Pro available as experimental options. The catch is disclosed up front. Users who switch into Forge consent to share their prompts, code, and the full sequence of edits, errors, and repairs the agent produces during a session, and those traces feed a training run with Arcee AI aimed at a trillion-parameter open-weight coding model. Teams and Enterprise workspaces are excluded from both Forge and the data collection pipeline.
This launch matters right now because closed-model pricing has been the practical cap on how much most solo builders can actually ship with agents like Bolt. Pairing a 50x allowance with a data-for-compute exchange gives individual developers meaningfully more room to prototype while routing real software-building traces (the kind public repos rarely capture) into an open-model lab. Alongside Forge, Bolt is also opening a waitlist for Bolt Lite, a $9 per month plan for students and side projects that includes the same Forge access, with signups closing October 14.
Comments (14)
Sign in to join the discussion.
P
Priya Ravindran29d ago
The 'and one more thing' at the end is very much cosplaying a Cupertino keynote. Respect the audacity.
M
Makoto K.29d ago
Model picker UI is finally breathing, but the little logo lockups for GLM and Kimi feel squished against the dropdown padding. Fix the vertical rhythm and it sings.
É
Étienne Marchand29d ago
The cut on that reveal video is tight but the voiceover mix is a hair too hot against the synth pad. Small nit on an otherwise crisp drop.
B
Bala Srinivasan29d ago
Free until Oct 14th with zero usage charges is a spicy way to torch some inference budget. Hope the CFO signed off before the tweet went live.
O
Olu Adebanjo29d ago
Cool, but is there a way to pin a default model per project via API? Right now I'm juggling env vars like it's 2021.
N
Noémi Balogh29d ago
Rate limits on the free tier? Or is 'up to 50x' the polite way of saying 'we'll throttle you at 3pm PT'?
M
Mara Kovač29d ago
Every 'up to 50x more usage' claim I've ever seen has done a lot of heavy lifting on the word 'up to'. Show me the median, not the ceiling.
D
Dmitri Volkov29d ago
Adding more models is nice but the whole 'pick a model' paradigm is already dead. In 6 months an agent will route this for you and nobody will remember what GLM stood for.
Z
Zahra Al-Mansouri29d ago
Been telling my portfolio for months that the coding tool wars are entering the 'give it away' phase. Popcorn ready.
H
Helga Lindqvist29d ago
Before I get excited: does Forge inherit the same SOC2 posture as the main product, or is this a separate trust boundary my security team will yell at me about?
X
Xiulan Deng29d ago
The launch tweet is doing numbers but the reply-to-quote ratio is suspiciously balanced. Either the community is very online or the notifications tab is working overtime.
K
kwamecrypto29d ago
GLM, DeepSeek, Kimi walk into a bar and nobody has launched a token for the router yet. Rare inefficiency.
J
Jun Park29d ago
Ok the Forge branding is genuinely good, annoyed I have to admit it. Building in this space and the naming bar keeps climbing.
F
Fintan O'Rourke29d ago
'Zero usage charges' is four words that made an entire finance team spit out their oat milk.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Bolt.new
Create stunning apps & websites by chatting with AI.
Create stunning apps & websites by chatting with AI.
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Bolt.new, the browser-based AI app builder from StackBlitz (led by co-founder and CEO Eric Simons), is rolling out Bolt Forge, a new agent that sits in the model picker alongside Standard and Max and runs on open-weight models instead of the closed frontier systems that power the rest of the product. Individual Pro subscribers get up to 50x their normal usage allowance on Forge at no extra cost through October 14, and when the allowance runs out the account rolls back to Standard rather than incurring overage charges. Under the hood, Forge routes work across GLM 5.3 Flash as the default, with GLM 5.3, Kimi K3, and DeepSeek v4 Pro available as experimental options. The catch is disclosed up front. Users who switch into Forge consent to share their prompts, code, and the full sequence of edits, errors, and repairs the agent produces during a session, and those traces feed a training run with Arcee AI aimed at a trillion-parameter open-weight coding model. Teams and Enterprise workspaces are excluded from both Forge and the data collection pipeline. This launch matters right now because closed-model pricing has been the practical cap on how much most solo builders can actually ship with agents like Bolt. Pairing a 50x allowance with a data-for-compute exchange gives individual developers meaningfully more room to prototype while routing real software-building traces (the kind public repos rarely capture) into an open-model lab. Alongside Forge, Bolt is also opening a waitlist for Bolt Lite, a $9 per month plan for students and side projects that includes the same Forge access, with signups closing October 14.
Comments (14)
The 'and one more thing' at the end is very much cosplaying a Cupertino keynote. Respect the audacity.
Model picker UI is finally breathing, but the little logo lockups for GLM and Kimi feel squished against the dropdown padding. Fix the vertical rhythm and it sings.
The cut on that reveal video is tight but the voiceover mix is a hair too hot against the synth pad. Small nit on an otherwise crisp drop.
Free until Oct 14th with zero usage charges is a spicy way to torch some inference budget. Hope the CFO signed off before the tweet went live.
Cool, but is there a way to pin a default model per project via API? Right now I'm juggling env vars like it's 2021.
Rate limits on the free tier? Or is 'up to 50x' the polite way of saying 'we'll throttle you at 3pm PT'?
Every 'up to 50x more usage' claim I've ever seen has done a lot of heavy lifting on the word 'up to'. Show me the median, not the ceiling.
Adding more models is nice but the whole 'pick a model' paradigm is already dead. In 6 months an agent will route this for you and nobody will remember what GLM stood for.
Been telling my portfolio for months that the coding tool wars are entering the 'give it away' phase. Popcorn ready.
Before I get excited: does Forge inherit the same SOC2 posture as the main product, or is this a separate trust boundary my security team will yell at me about?
The launch tweet is doing numbers but the reply-to-quote ratio is suspiciously balanced. Either the community is very online or the notifications tab is working overtime.
GLM, DeepSeek, Kimi walk into a bar and nobody has launched a token for the router yet. Rare inefficiency.
Ok the Forge branding is genuinely good, annoyed I have to admit it. Building in this space and the naming bar keeps climbing.
'Zero usage charges' is four words that made an entire finance team spit out their oat milk.