Fireworks(@FireworksAI_HQ)

The MiniMax M3 model is now available for training on Fireworks! You can use managed LoRA SFT and D...

Fireworks AI (@FireworksAI_HQ) — Tue, 21 Jul 2026 23:51:07 GMT

The MiniMax M3 model is now available for training on Fireworks!

You can use managed LoRA SFT and DPO for standard fine-tuning workflows, or the Fireworks Training API for custom SFT, DPO, and RL loops with checkpointing, rollout inference, and adapter hotloading.

Get started:
app.fireworks.ai/models/firewor…github.com/fw-ai/cookbook…

💬7🔄4❤️99👀6892📊21

Full details https://t.co/bFj6TKhaTa

Fireworks AI (@FireworksAI_HQ) — Tue, 21 Jul 2026 18:34:26 GMT

Full details fireworks.ai/blog/kimik3-fa…

💬0🔄7❤️107👀34204📊35

We ran Kimi K3 against Fable on ~1,000 agentic tasks, expecting a catch-up story. We got a specializ...

Fireworks AI (@FireworksAI_HQ) — Tue, 21 Jul 2026 18:34:02 GMT

We ran Kimi K3 against Fable on ~1,000 agentic tasks, expecting a catch-up story. We got a specialization story instead.

@kimi_moonshot's K3 outperformed on security, crypto, and long terminal loops. Fable beat on multi-lang + web/data viz. Per-task routing hits 93% accuracy, above BOTH models, at up to 50x lower cost than Fable on long loops.

The part nobody's pricing in yet: the router sends 72-96% of traffic to K3. The frontier model becomes the fallback rather than the default.

Kimi K3, coming to Fireworks July 27.

💬62🔄185❤️2098👀495969📊595

The team @tryheidi didn't want to keep renting someone else's intelligence. So Heidi fine-tuned an ...

Fireworks AI (@FireworksAI_HQ) — Tue, 21 Jul 2026 18:26:16 GMT

The team @tryheidi didn't want to keep renting someone else's intelligence.

So Heidi fine-tuned an open model that beat Gemini Pro on quality in their internal evals, and ran with 3.5x faster latency (25s → 7s). From POC to production in 4 weeks.

Get the full scoop → fireworks.ai/blog/Heidi-Hea…

💬0🔄2❤️53👀4942📊10

We'll see you in an hour!

Fireworks AI (@FireworksAI_HQ) — Tue, 21 Jul 2026 16:01:29 GMT

We'll see you in an hour!

Sinan Ozdemir@Prof_OZ

Want to chat live with me about fine-tuning, Kimi K3, or really anything else AI?

Come to our first of many @FireworksAI_HQ office hours tomorrow @ 10am PT

See you there!
fireworks.ai/event/office-h…

💬0🔄0❤️11👀4506📊1

Our first DevRel office hours kicks off tomorrow. Come chat with our Head of AI Education @Prof_OZ a...

Fireworks AI (@FireworksAI_HQ) — Mon, 20 Jul 2026 20:14:04 GMT

Our first DevRel office hours kicks off tomorrow. Come chat with our Head of AI Education @Prof_OZ about anything and everything AI.

Sinan Ozdemir@Prof_OZ

Want to chat live with me about fine-tuning, Kimi K3, or really anything else AI?

Come to our first of many @FireworksAI_HQ office hours tomorrow @ 10am PT

See you there!
fireworks.ai/event/office-h…

💬1🔄1❤️15👀3696📊3

Open-weight models like GLM-5.2 and Kimi K2.6 demonstrate the capability & performance needed fo...

Fireworks AI (@FireworksAI_HQ) — Fri, 17 Jul 2026 17:04:37 GMT

Open-weight models like GLM-5.2 and Kimi K2.6 demonstrate the capability & performance needed for real GTM work. 

@ClayRunHQ's team recognized that early, and we're proud to be the inference layer making it real.

Congrats to @jeffbarg and Clay. Let's keep building together.

Jeff Barg@jeffbarg

x.com/i/article/2077…

💬2🔄0❤️19👀5071📊4

It has been an incredible day, thank you all for celebrating with us. If you'd like to come build ...

Fireworks AI (@FireworksAI_HQ) — Thu, 16 Jul 2026 23:24:03 GMT

It has been an incredible day, thank you all for celebrating with us. 

If you'd like to come build with us, we can't wait to meet you: fireworks.ai/careers

This is only the beginning.

Pranav Jain@pranavj09

Thanks to all our incredible customers and partners for being part of this journey.

We’re entering an era of specialized intelligence, where every company will build models tailored to its users that continuously improve over time. It’s time to own your intelligence and control your destiny.

If you’re building the future of AI, we’d love to build with you.

The future of AI is specialized. We’re just getting started.

💬2🔄0❤️33👀7074📊5

Our CEO @lqiao will be joining the live broadcast with @MTSlive at 2:30pm PT to talk about building ...

Fireworks AI (@FireworksAI_HQ) — Thu, 16 Jul 2026 18:46:19 GMT

Our CEO @lqiao will be joining the live broadcast with @MTSlive at 2:30pm PT to talk about building specialized intelligence, and our $1.5B Series D and $1B ARR milestone.

MTS@MTSlive

x.com/i/broadcasts/1…

💬0🔄1❤️14👀4186📊2

Grateful to @BessemerVP for being an early believer. The curve isn't really ours to claim. It belong...

Fireworks AI (@FireworksAI_HQ) — Thu, 16 Jul 2026 17:32:06 GMT

Grateful to @BessemerVP for being an early believer. The curve isn't really ours to claim. It belongs to every builder who choses to own their intelligence instead of renting it. We just built the platform to make it possible.

Bessemer@BessemerVP

𝐅𝐢𝐫𝐞𝐰𝐨𝐫𝐤𝐬 𝐠𝐫𝐞𝐰 𝐟𝐫𝐨𝐦 $𝟏𝟎𝟎𝐌 𝐀𝐑𝐑 𝐭𝐨 $𝟏𝐁 𝐀𝐑𝐑 𝐢𝐧 𝟏𝟔 𝐦𝐨𝐧𝐭𝐡𝐬!

@AnthropicAI set the revenue trajectory for the next generation of AI Giants. And @FireworksAI_HQ demonstrates the path to $1B ARR in 4 years, isn’t a fluke, but a new possibility.

𝐅𝐢𝐫𝐞𝐰𝐨𝐫𝐤𝐬 (𝐉𝐚𝐧 𝟐𝟎𝟐𝟐) 𝐯𝐬. 𝐀𝐧𝐭𝐡𝐫𝐨𝐩𝐢𝐜 (𝐉𝐚𝐧 𝟐𝟎𝟐𝟏)

— 𝐅𝐨𝐮𝐧𝐝𝐢𝐧𝐠 →  $𝟏𝟎𝟎𝐌 𝐀𝐑𝐑: Both took ~3 years

— $𝟏𝟎𝟎𝐌 → $𝟏𝐁 𝐀𝐑𝐑: Fireworks (16 months) vs. Anthropic (11-12 months)

— 𝐅𝐨𝐮𝐧𝐝𝐢𝐧𝐠 → $𝟏𝐁 𝐀𝐑𝐑: Fireworks (~4 years) vs. Anthropic (~4 years)

💬0🔄2❤️64👀9373📊8

Our Fine-tuning 101 webinar is just about to start, you still have time to jump in: https://t.co/ddc...

Fireworks AI (@FireworksAI_HQ) — Thu, 16 Jul 2026 17:02:26 GMT

Our Fine-tuning 101 webinar is just about to start, you still have time to jump in: fireworks.ai/event/fine-tun…

Can't make it this time? That's alright, we'll post the webinar on YouTube this week!

Fireworks AI@FireworksAI_HQ

Our free monthly DevRel webinar series kicks off this Thursday, 10am PST.

Fine-tune an open vision model to extract clean, structured JSON from messy receipt images, managed start to finish on Fireworks. 

Plus recent releases and a fine-tuning primer.

fireworks.ai/event/fine-tun…

💬2🔄0❤️13👀2965📊3

Every company must own its intelligence. We've raised $1.5B Series D at a $17.5B valuation. We’ve...

Fireworks AI (@FireworksAI_HQ) — Thu, 16 Jul 2026 13:26:18 GMT

Every company must own its intelligence.

We've raised $1.5B Series D at a $17.5B valuation. 

We’ve surpassed $1B ARR and serve over 40 trillion tokens daily, with 95%+ coming from models specialized on customer data.

We’re just getting started.

More: fireworks.ai/blog/series-d-…

Your browser does not support the video tag.

💬54🔄54❤️914👀314697📊240

Generic AI models are not built for medicine. Doximity Ask, trained and served on Fireworks, outper...

Fireworks AI (@FireworksAI_HQ) — Wed, 15 Jul 2026 22:35:30 GMT

Generic AI models are not built for medicine.

Doximity Ask, trained and served on Fireworks, outperformed GPT-5.6 Sol, Claude Fable 5, and OpenEvidence in a Stanford-Harvard clinical AI safety study.

Specialized intelligence beats the generic version when safety matters. @doximity proved it.

Doximity@doximity

Finally: an independent, head-to-head test of clinical AI models.  

This week, Stanford's Arise Lab published one of the most comprehensive independent evaluations of clinical AI to date. The benchmark compared 24 frontier and clinical AI models, producing 12,747 expert rankings on their answers to real-world clinical questions.  

We were excited to see Doximity Ask outperform 
OpenEvidence, GPT-5.6, Claude Fable 5, Gemini 3.1 Pro, and other leading frontier AI models.  

The bigger story, though, is that clinical AI is finally getting the kind of rigorous, independent benchmarking it needs.

💬2🔄3❤️29👀6904📊7

The best base model for training isn't about someone else's benchmarks. Excited to offer @thinkymac...

Fireworks AI (@FireworksAI_HQ) — Wed, 15 Jul 2026 22:08:45 GMT

The best base model for training isn't about someone else's benchmarks.

Excited to offer @thinkymachines' first open-weights model: Inkling. 975B MoE, multimodal, with Apache 2.0. A great foundation for fine-tuning. Available day-zero.

fireworks.ai/models/firewor…

💬2🔄4❤️106👀10408📊18

Trending AND fastest-growing in the same month? Benchmarks change daily, but only @tryramp has the d...

Fireworks AI (@FireworksAI_HQ) — Wed, 15 Jul 2026 20:43:44 GMT

Trending AND fastest-growing in the same month? Benchmarks change daily, but only @tryramp has the database of real receipts to track this. We can confirm: demand for open-weight inference and training is on fire🔥. Congrats to @HiggsField_AI landing on both lists too.

Ara Kharazian@arakharazian

New: Top SaaS Vendors on @tryramp July 2026

Two trends this month:

1. AI support bots are one of the clearest enterprise AI use cases showing up in payments data (PolyAI, Sierra)

2. More and more usage of open source and Chinese models (DeepSeek, Fireworks AI, Baseten, Together, OpenRouter)

💬1🔄0❤️16👀3596📊3

He's right, you know.

Fireworks AI (@FireworksAI_HQ) — Wed, 15 Jul 2026 00:11:33 GMT

He's right, you know.

Ara@arafatkatze

Turns out you can get an opus 4.8 level model perf at 190 tokens per second and 1/4th of the price if you use an Open weights model called GLM-5.2 Fast.

💬5🔄0❤️29👀4906📊7

Last month we hosted an AI Nerd Meetup @Workato HQ featuring builders tackling the hardest problems...

Fireworks AI (@FireworksAI_HQ) — Tue, 14 Jul 2026 22:39:00 GMT

Last month we hosted an AI Nerd Meetup @Workato HQ  featuring builders tackling the hardest problems in agentic AI.

Highlights:
@yuan_sihan of Fireworks - how open-source agents can use a frontier model as a reviewer to hit much higher quality without paying frontier costs at every step. 

The research behind our latest post, "Frontier AI at a fraction of the cost": fireworks.ai/blog/frontier-…@KovacVeljko of @cognee_ - fact disambiguation and agent memory: how agents retain, retrieve, and use info over time, and why open-source memory infra matters in production.

@mat_pichler of @wordware - collaborative filesystems for agents: how shared file and workspace layers help agents coordinate, persist work, and collaborate.

Thanks to our speakers and Workato for hosting!

Your browser does not support the video tag.

💬3🔄1❤️12👀2265📊6

Our free monthly DevRel webinar series kicks off this Thursday, 10am PST. Fine-tune an open vision ...

Fireworks AI (@FireworksAI_HQ) — Tue, 14 Jul 2026 16:34:06 GMT

Our free monthly DevRel webinar series kicks off this Thursday, 10am PST.

Fine-tune an open vision model to extract clean, structured JSON from messy receipt images, managed start to finish on Fireworks. 

Plus recent releases and a fine-tuning primer.

fireworks.ai/event/fine-tun…

💬1🔄5❤️29👀249409📊7

ICYMI: @LangChain Deep Agents on @nvidia Nemotron 3 Ultra. frontier open-model agents at ~10x lower ...

Fireworks AI (@FireworksAI_HQ) — Tue, 14 Jul 2026 14:51:38 GMT

ICYMI: @LangChain Deep Agents on @nvidia Nemotron 3 Ultra. frontier open-model agents at ~10x lower cost than closed. 

Run on Fireworks, then post-train it into specialized intelligence you own.

fireworks.ai/blog/Open-fron…

💬1🔄2❤️17👀3298📊5

Our new devrel webinar series kicks off July 16th with our new Head of AI Education @Prof_OZ ↓

Fireworks AI (@FireworksAI_HQ) — Sun, 12 Jul 2026 14:12:08 GMT

Our new devrel webinar series kicks off July 16th with our new Head of AI Education @Prof_OZ ↓

Sinan Ozdemir 🎆@Prof_OZ

I'm launching a live series at @FireworksAI_HQ to help you ship on inference + training

Every month:
⚡Webinar - live demos + what's new at Fireworks
⚡2 office hours - bring any AI/Fireworks question
⚡Discord AMA

First webinar runs next Thursday 👇fireworks.ai/event/fine-tun…0

💬4🔄2❤️13👀4694📊7

Long-context sparse attention has a catch: data-dependent block selection wrecks memory access kills...

Fireworks AI (@FireworksAI_HQ) — Fri, 10 Jul 2026 23:21:35 GMT

Long-context sparse attention has a catch: data-dependent block selection wrecks memory access kills speed.

Our @MiniMax_AI M3 kernel on Blackwell answers it. KV-stationary, each block read once, ~980 TFLOP/s on a B200.

See the breakdown here →
fireworks.ai/blog/kernel-op…

💬7🔄5❤️73👀33375📊22

Loving this piece: bringing delight to your daily coffee, your daily workflows, and your AI bill. ...

Fireworks AI (@FireworksAI_HQ) — Thu, 09 Jul 2026 17:28:31 GMT

Loving this piece: bringing delight to your daily coffee, your daily workflows, and your AI bill. 

@gumloop is now running open-weight models on Fireworks in production. 80% lower cost vs. closed-lab APIs. 

Disbelief might be the reaction. But the savings are real.

Max Brodeur-Urbas@MaxBrodeurUrbas

there are certain costs people have stopped questioning

so when one of them drops 80%, the first reaction is disbelief

open-weight models will give everyone that reaction at scale this year

excited to be partnering with @FireworksAI_HQ on this

Your browser does not support the video tag.

💬6🔄1❤️33👀3627📊9

Not another demo or benchmark. @ShoucongChen is a senior member of our technical staff. A real pro...

Fireworks AI (@FireworksAI_HQ) — Wed, 08 Jul 2026 20:08:29 GMT

Not another demo or benchmark. @ShoucongChen is a senior member of our technical staff. 

A real project, scoped at 1-month. Delivered in 4 days with GLM5.2 Fast.  

The best devs deserve >400 t/sec. 

Take it for a spin in Claude Code via FireConnect and tell us what you think.

Shoucong Chen@ShoucongChen

For complex SWE projects, speed matters more than I thought: slow agents make me context switch.
GLM 5.2 Fast runs at ~400 tok/s, finishing before I need to switch tabs.
A ~1 engineer-month task took 4 days and $218 in inference.
Story: fireworks.ai/blog/glm5p2-fa…@FireworksAI_HQ

💬5🔄4❤️35👀55614📊10

If you're tired of re-explaining the same context to your AI every day, go see what @PromptQL just l...

Fireworks AI (@FireworksAI_HQ) — Wed, 08 Jul 2026 17:55:27 GMT

If you're tired of re-explaining the same context to your AI every day, go see what @PromptQL just launched!

Tanmai Gopal@tanmaigo

We raised $136M to kill Slack.

Introducing PromptQL: The first AI version of Slack.

Here’s how it works:

Your browser does not support the video tag.

💬7🔄2❤️36👀6292📊11

This is what production RL infra looks like. ICYMI: RL training at scale separates into two distinc...

Fireworks AI (@FireworksAI_HQ) — Wed, 08 Jul 2026 17:05:28 GMT

This is what production RL infra looks like.

ICYMI: RL training at scale separates into two distinct problems.
- Tight collective comms for the trainer.
- Distributed async inference for rollout.

Kudos to the @cognition team on this. Their trainer is the secret sauce behind SWE-1.7 and reliable RL rollouts across four datacenters on three continents is easier said than done!

We're excited to continue partnering with AI leaders on their path to specialized intelligence.

Cognition@cognition

Our RL training spans four datacenters across three continents, combining our own GPUs across multiple clusters with additional compute from inference providers like @FireworksAI_HQ.

Only the trainer needs tight collective communication. Rollout inference was distributed, with engines syncing through compressed weight diffs in object storage.

💬3🔄8❤️91👀158262📊16

Attending @RaiseSummit? Join our CEO @lqiao tomorrow @ 4:00 PM for a fireside chat with @MattEvant...

Fireworks AI (@FireworksAI_HQ) — Tue, 07 Jul 2026 17:49:28 GMT

Attending @RaiseSummit? 

Join our CEO @lqiao tomorrow @ 4:00 PM for a fireside chat with @MattEvantic of Evantic Capital on the Master Stage. 

See you in Paris!

💬0🔄2❤️8👀2344📊2

What actually makes an AI application good or not? Thanks to @StackOverflow for hosting @the_bunny_...

Fireworks AI (@FireworksAI_HQ) — Tue, 07 Jul 2026 16:57:36 GMT

What actually makes an AI application good or not?

Thanks to @StackOverflow for hosting @the_bunny_chen on the Stack Overflow Podcast (even though he is giving away all of our secrets!)

Listen here: stackoverflow.blog/2026/07/03/the…

Stack Overflow@StackOverflow

🎙️ We welcome Benny Chen @the_bunny_chen, co-founder of @FireworksAI_HQ, to the show to explore what actually makes an AI application good or not, how to balance qualitative signals with quantitative metrics when evaluating AI, and how open-source eval protocols and community efforts are setting the standard for AI evaluation.
stackoverflow.blog/2026/07/03/the…u

Your browser does not support the video tag.

💬0🔄0❤️7👀3087📊1

Today, we're helping kick off AMD AI Developer Hackathon: ACT II. We're giving $50 in Fireworks cr...

Fireworks AI (@FireworksAI_HQ) — Mon, 06 Jul 2026 16:52:37 GMT

Today, we're helping kick off AMD AI Developer Hackathon: ACT II. 

We're giving $50 in Fireworks credits to participants in this event which also includes:

$20K in prizes. $150 in credits. Fully online.

Powered by @AIatAMD@lablabai 

Sign up → lablab.ai/ai-hackathons/…

Your browser does not support the video tag.

💬4🔄4❤️50👀6086📊17

Don't forget, we'll be in Paris for @RaiseSummit. Will we see you there?

Fireworks AI (@FireworksAI_HQ) — Fri, 03 Jul 2026 16:41:54 GMT

Don't forget, we'll be in Paris for @RaiseSummit.

Will we see you there?

Fireworks AI@FireworksAI_HQ

Going to be in Paris for @RaiseSummit? 

Join us and @nvidia to hear from leaders at both companies on the future of AI inference and infrastructure.

After that: cocktails and a DJ. Oh la la! 

See you there: luma.com/no98hixu

💬1🔄0❤️16👀3020📊2

Fireworks Batch API: 50% cheaper than serverless. We obviously love things fast, but sometimes asy...

Fireworks AI (@FireworksAI_HQ) — Wed, 01 Jul 2026 23:31:12 GMT

Fireworks Batch API: 50% cheaper than serverless. 

We obviously love things fast, but sometimes async at scale is all you need. With the refreshed Batch API, you queue up a job and select whether you need it completed in 12/24/48/72 hours. Plus automatic prompt caching for even more savings.

Submit a JSONL and walk away. Go enjoy some soccer or fireworks, and come back to results. 🎆

Get started now →docs.fireworks.ai/guides/batch-i…H

💬6🔄4❤️43👀150026📊15

If you want frontier-level coding and agent performance but you don't want to pay closed-model price...

Fireworks AI (@FireworksAI_HQ) — Wed, 01 Jul 2026 23:10:34 GMT

If you want frontier-level coding and agent performance but you don't want to pay closed-model prices, GLM 5.2 is the open model you've probably been hearing about. 

Here's why you should run it on Fireworks.

Your turn: fireworks.ai/models/firewor…

Your browser does not support the video tag.

💬9🔄2❤️62👀145465📊19

Frontier open models with enterprise governance means rebuilding workflows from scratch. Move faster...

Fireworks AI (@FireworksAI_HQ) — Wed, 01 Jul 2026 19:51:11 GMT

Frontier open models with enterprise governance means rebuilding workflows from scratch. Move faster with Fireworks on Foundry.

GLM 5.2 is live on Microsoft Foundry. 

With FireConnect enabled, devs can use their Foundry PTUs to route GLM through Codex, OpenCode, or Pi. 

Get started today, and let us know what you think →
docs.fireworks.ai/ecosystem/fire…

💬3🔄3❤️50👀4049📊11

Inference reliability has historically been a tax on devs that only large well-funded startups could...

Fireworks AI (@FireworksAI_HQ) — Wed, 01 Jul 2026 00:09:18 GMT

Inference reliability has historically been a tax on devs that only large well-funded startups could afford: reserve GPUs in advance, sign a contract, guess your peak throughput requirements. 

Everyone else has been at the mercy of the market, and deals with the occasional 503s and rate limits. 

Serverless 2.0 flips that: same production grade reliability you'd get with a dedicated deployment, and you only pay the premium for priority tier when you need it.

elvis@omarsar0

x.com/i/article/2071…

💬5🔄4❤️51👀11511📊14

For even higher speeds, reach out for a custom deployment! We’ve hit 446 tok/s on Artificial Analys...

Fireworks AI (@FireworksAI_HQ) — Tue, 30 Jun 2026 19:29:20 GMT

For even higher speeds, reach out for a custom deployment!

We’ve hit 446 tok/s on Artificial Analysis.

Learn more → fireworks.ai/blog/glm-5p2-f…

💬0🔄0❤️21👀2198📊3

We heard your feedback. You want to go faster. Introducing GLM 5.2 Fast The same model and quality...

Fireworks AI (@FireworksAI_HQ) — Tue, 30 Jun 2026 19:29:19 GMT

We heard your feedback. You want to go faster.

Introducing GLM 5.2 Fast

The same model and quality as GLM 5.2 standard, now at 140 tok/s

Flip one model ID → accounts/fireworks/routers/glm-5p2-fast

💬14🔄16❤️309👀361457📊68

Going to be in Paris for @RaiseSummit? Join us and @nvidia to hear from leaders at both companies ...

Fireworks AI (@FireworksAI_HQ) — Mon, 29 Jun 2026 15:55:35 GMT

Going to be in Paris for @RaiseSummit? 

Join us and @nvidia to hear from leaders at both companies on the future of AI inference and infrastructure.

After that: cocktails and a DJ. Oh la la! 

See you there: luma.com/no98hixu

💬5🔄3❤️23👀264574📊9

Model management is the real SDLC scaling bottleneck. @FactoryAI standardized on Fireworks to solv...

Fireworks AI (@FireworksAI_HQ) — Sat, 27 Jun 2026 00:19:52 GMT

Model management is the real SDLC scaling bottleneck.

@FactoryAI  standardized on Fireworks to solve it: 
→ 2–3x open-model growth 
→ 5–15x more work per dollar
→ day-0 access to every new open-weight model in production.

Watch Leo Tchourakov break it down → youtube.com/watch?v=7ZaKJz…

💬3🔄1❤️26👀4534📊7

Read the full case study here: Full case study: https://t.co/BOVDPIURML

Fireworks AI (@FireworksAI_HQ) — Sat, 27 Jun 2026 00:19:52 GMT

Read the full case study here: Full case study: fireworks.ai/blog/Factory

💬1🔄0❤️1👀934📊1

The big lesson from training @cursor_ai Composer 2: models exploit flaws in their training environme...

Fireworks AI (@FireworksAI_HQ) — Fri, 26 Jun 2026 23:06:04 GMT

The big lesson from training @cursor_ai Composer 2: models exploit flaws in their training environment before learning what you actually want. 

Real RL for coding agents means production-faithful environments + distributed infra to match.

Great breakdown from @ellev3n11 and @dzhulgakov 

Read more: fireworks.ai/blog/Cursor-Co…

Your browser does not support the video tag.

💬4🔄6❤️45👀5352📊14

We hosted the first RSI RL Environments hackathon with @hud_evals @ycombinator and it was a blast! ...

Fireworks AI (@FireworksAI_HQ) — Fri, 26 Jun 2026 16:29:16 GMT

We hosted the first RSI RL Environments hackathon with @hud_evals@ycombinator and it was a blast!

We watched builders treat RL as a general purpose tool and reach for it across domains we had not anticipated.

Teams were building for chip design, gaming, robotics, and more.

💬2🔄2❤️17👀2106📊5

RL fine-tuning is now live for @nvidiaai Nemotron 3 on Fireworks, starting with Nemotron 3 Super (Lo...

Fireworks AI (@FireworksAI_HQ) — Thu, 25 Jun 2026 23:54:09 GMT

RL fine-tuning is now live for @nvidiaai Nemotron 3 on Fireworks, starting with Nemotron 3 Super (LoRA). Train with GRPO and serve the model in one place. 

We price by GPU-hour, not per token, so long multi-turn rollouts don't blow up your bill.

Training shapes → docs.fireworks.ai/fine-tuning/tr…

💬0🔄1❤️24👀2791📊4

In a joint Fireworks and @Faros_AI evaluation of 211 real engineering tasks, Claude Code + GLM-5.2 b...

Fireworks AI (@FireworksAI_HQ) — Thu, 25 Jun 2026 16:26:41 GMT

In a joint Fireworks and @Faros_AI evaluation of 211 real engineering tasks, Claude Code + GLM-5.2 beat both Claude Code + Opus 4.8 and Codex + GPT-5.5:

- Judge score: 0.568 vs. 0.521 and 0.466
- Time per task: 321s vs. 775s and 392s
- Cost per task: $0.92 vs. $1.76 and $2.06

Most importantly, Faros tested the models on its own repositories and work, not just public benchmarks.

Model choice should be based on where a model clears your bar, on your work, at a cost that makes sense.

Faros@Faros_AI

With our partner @fireworksai_hq, we ran an evaluation to find out if open models have gotten strong enough to handle real software engineering work. Our result: Claude Code + GLM-5.2 matched Kimi on quality while running faster and cheaper. More details:
faros.ai/blog/open-mode…

💬8🔄14❤️163👀17132📊32

Open + closed models = better together. Our previous research with @harvey showed the benefits of c...

Fireworks AI (@FireworksAI_HQ) — Thu, 25 Jun 2026 04:19:37 GMT

Open + closed models = better together.

Our previous research with @harvey showed the benefits of combining a frontier closed model as an advisor agent with fine-tuned, open-source worker agents.  Three benchmarks later, the result holds. In an even simpler set up, the open + closed duo delivers superior outcomes and 40-67% lower cost than running entirely on Opus 4.8.

Read the breakdown → fireworks.ai/blog/frontier-…

💬6🔄14❤️116👀16949📊43

It's way easier to switch models than to switch harnesses, and like many of you we use @cursor_ai e...

Fireworks AI (@FireworksAI_HQ) — Wed, 24 Jun 2026 23:02:16 GMT

It's way easier to switch models than to switch harnesses, and like many of you we use @cursor_ai  every day.

Now you can try out the latest open-source frontier model without changing your workflow.

Lee Robinson@leerob

You can now try GLM 5.2 in Cursor!

Excited to see more useful open models, thank you to Fireworks for partnering here. Results from our evals ↓

💬5🔄4❤️78👀7777📊12

The hard part of reinforcement learning on a frontier model is the infrastructure that keeps trainin...

Fireworks AI (@FireworksAI_HQ) — Wed, 24 Jun 2026 20:00:58 GMT

The hard part of reinforcement learning on a frontier model is the infrastructure that keeps training and inference numerically identical: zero KLD, end to end. 

We've solved this challenge, and are now offering it as a managed service, starting with GLM 5.2.

Benny (Yufei) Chen@the_bunny_chen

x.com/i/article/2069…

💬4🔄20❤️68👀12733📊21

GLM 5.2 is the open coding model everyone's been talking about. Now you can fine-tune it on Firework...

Fireworks AI (@FireworksAI_HQ) — Wed, 24 Jun 2026 19:07:59 GMT

GLM 5.2 is the open coding model everyone's been talking about. Now you can fine-tune it on Fireworks. SFT, DPO, and RL all supported.

A leaderboard winner can still lose on your codebase. Training closes that gap. 

Take the strongest open coding base and turn it into a model that's tuned to your evals, your prompts, your latency requirements. 

And what you train serves on the same stack in production. No handoff, no migration.

We opened private access to select customers over the weekend. Early reaction has been 🔥

What are you waiting for:docs.fireworks.ai/api-reference/…z

💬5🔄1❤️52👀3610📊14

Bring open models to where you're already working. FireConnect brings Fireworks' top models directl...

Fireworks AI (@FireworksAI_HQ) — Tue, 23 Jun 2026 22:59:56 GMT

Bring open models to where you're already working.

FireConnect brings Fireworks' top models directly into Claude Code, Pi, OpenCode, and Codex.

Watch our Head of AI Education @Prof_oz show you how:

Sinan Ozdemir 🎆@Prof_OZ

How to run GLM-5.2 (Or Minimax or Qwen or Deepseek or Kimi) in Claude Code

No JSON editing. No vim 😮‍💨

1. Install Claude Code
npm install -g @anthropic-ai/claude-code

2. Grab a Fireworks API key fireworks.ailC

3. Install FireConnect + drop in your key (one line):
curl -fsSraw.githubusercontent.com/fw-ai/fireconn…99 | FIREWORKS_API_KEY="fw_..." bash

4. Flip it on:
 fireconnect claude on

5. Switch to GLM-5.2 and bam! Swap models with:
fireconnect claude model select

Docs to do the same for Pi, Opencode, and Codexdocs.fireworks.ai/ecosystem/fire…IZ

Your browser does not support the video tag.

💬3🔄5❤️34👀142249📊9

You don't need to be AI Dependent to build State of the Art software. @FactoryAI and Fireworks use...

Fireworks AI (@FireworksAI_HQ) — Mon, 22 Jun 2026 20:59:33 GMT

You don't need to be AI Dependent to build State of the Art software. 

@FactoryAI and Fireworks use GLM to build the most advanced agentic AI you can train and own today.

Factory@FactoryAI

GLM 5.2 is available in Droid, hosted by @FireworksAI_HQ.

💬8🔄2❤️53👀5720📊13

GLM-5.2 has been the most popular new model on Fireworks this past week. @ArtificialAnlys confirms ...

Fireworks AI (@FireworksAI_HQ) — Mon, 22 Jun 2026 19:59:56 GMT

GLM-5.2 has been the most popular new model on Fireworks this past week.

@ArtificialAnlys confirms why: #3 overall on GDPval-AA (1524 Elo), #1 open weights by 116 points.

Interest is showing no signs of slowing.

Try it on Fireworks today → fireworks.ai/models/firewor…

Artificial Analysis@ArtificialAnlys

GLM-5.2 leads open weights models and sits at #3 overall on GDPval-AA, a real-world agentic work benchmark

GLM-5.2 from @Zai_org scores 1524 Elo on GDPval-AA, which measures performance on real-world, economically valuable knowledge work through long-horizon, multi-turn tasks.

Key takeaways:

➤ #3 overall, behind only Claude Fable 5 (1783) and Claude Opus 4.8 (1615), and level with GPT-5.5 (xhigh, 1509)

➤ The leading open weights model by a wide margin: the next open model, MiniMax-M3, scores 1408

➤ Ahead of many proprietary models, including Google's Gemini 3.5 Flash (1357), Qwen 3.7 Max (1289), Muse Spark (1158)

➤ The tasks are agentic. GLM-5.2 averaged ~31 turns per task across 1,999 matches

➤ Consistent with the rest of its launch, GLM-5.2 also leads open weights on the Artificial Analysis Intelligence Index, ranks #3 on the Agentic Index, and #3 on AA-Briefcase