Next up in the AnsemHack Spotlight: Byte Arena @bytearenafun ships games and their agents are already in it, playing it, testing it and narrating it live to Twitch, X and pump.fun. Solana Survivors Season 2 has landed: Heaven or Hell.

Today we're spotlighting a late-comer to the hackathon: ForgeBench. @ForgeAI_gg is a protocol for pressure-testing agents against realtime games. Daily Dungeons, Counter-Strike, GTA V: every run gets a budget, a clock, and mistakes that stay made.

What happens when the fastest way to pay someone on Solana meets the fastest way to pay for inference?

Today’s AnsemHack highlight is on SelfMade by SP3ND @SP3NDdotshop isa tokenized agent that uses creator fees to buy GPUs. It has earned and transferred 938 SOL, roughly $106K, with $40K spent so far on machines through SP3ND’s own stablecoin shopping platform.

This afternoon’s project spotlight: MONARK @usemonark is building from the ground up a novel approach to AI in finance: one engine, two sides. For AI agents, a gate that decides if a decision can be trusted: go, wait, or don't. For DeFi, apps build on that same engine.

Today’s AnsemHack Spotlight is covering Market Bubble Search @mbubbleSearch is a search engine for crypto podcasts/broadcast. 679 episodes and 1,134 hours: Market Bubble, MCG Live, Elon's interviews. Ask in plain english, get the quote and the second it was said in ranked order.

This afternoon, we're spotlighting Corine @corinedotin is an agentic finance network on Solana where users can deploy autonomous financial agents from a prompt. Agents can monitor markets, execute onchain, remember outcomes and improve over time without users needing to write code.

This morning's AnsemHack spotlight is on ANSEM.TIPS ANSEM.TIPS (@tipansem) is a social tipping layer on Solana that lets anyone send crypto and tokenized stocks directly on X. No wallet address needed--your X username acts as your social address.

This afternoon's AnsemHack spotlight is on BUZZ @eienel_eth is a survival-pot game where humans and AI agents play the same hidden board: the comb with the fewest players is the one that dies, and half the pot pays out on prediction skill. 200 finished games on devnet.

MiMo-V2.6-Flash and MiMo-V2.6-Pro are live on UsePod The models are priced at $0.14/$0.28 (Flash) and $0.435/$0.87 (Pro) per million tokens. Released Today by Xiaomi with open weights. 1.02T total params / 42B active. Frozen-router MoE architecture. 1M context, 128K output tokens, multimodal (text, image, video, audio).

This morning's AnsemHack spotlight is Claw Hunter @clawhuntersol pairs a creative AI studio with a model gateway: think Higgsfield for making images and videos, plus OpenRouter for powering apps + agents. Studios by Claw Hunter and lets communities launch their own AI studios.

This afternoon, we chose to spotlight HANSEM @ansemdev gives every wallet a free hosted trading agent on Solana, then wraps it in a bigger stack: an agent-only social network, a token launcher, and 121 paid data skills with published per-call prices.

Next in our spotlight series is Peng Town! @penguinxbt_ is a cozy multiplayer world blending Club Penguin’s social charm with creature collecting and battling. Catch companions, build your team, decorate your home, and hang out with friends onchain in a persistent pixel world.

Next hackathon spotlight: Mizuki the Maintainer @MizukiMech is solving a coordination problem, not just a cost problem. It does fixed-price GitHub maintenance: label an issue, get a quote pinned to that commit, pay 2 USDC for up to 3 changed files or 10 USDC for 10 over x402.

Continuing our spotlight series, POD Miner. @pod_miner rents GPU boxes on Vast.ai, serves an open 8B model from them, and sells that capacity on the inference market; the spread is meant to buy and burn $PODM. A live console shows the box, the bond and the meter.

Next up in our spotlight is a project that needs no introduction, Hell's Agents! @HellsAgents is a live Solana betting game where AI riders race motorcycles for real SOL: five seats, ten fuel units spread across ten segments, one seed revealed only after every commit locks. Races are anchored on-chain. $HELLS: ~$10K cap, graduated.

Project spotlight #3 is Anima. Anima (@AnimaAgent) is a Solana research-and-swap agent with a 3D face and voice: it streams answers, runs live market tools, and returns unsigned swaps sealed in expiring, wallet-bound quote tokens. Fund the agent, share it, no seed phrase.

Next in our spotlight series is Grainlify. @Grainlify runs grants for open-source work: a funder escrows a pool, issues go to a published weighted draw instead of first-come, and contributors claim payouts with Merkle proofs. All 102 allocation rules are public.

First up in the AnsemHack spotlight is Hyre! @Hyre_agent sells pay-per-call inference and data to agents on Solana: 54 models through a single OpenAI-compatible gateway settled in USDC over x402. solana:2HtyE1W7fE2cdoYxR5fsukr9AyoQGpJep14NAVBMpump: $93K cap, $12.8K/24h.

As the AnsemHack Clawrena hackathon comes to a close, we'll be spotlighting projects building on UsePod on the Inference Marketplace track. 15% of the solana:9cRCn9rGT8V2imeM2BaKs13yhMEais3ruM3rPvTGpump prize pool + $10,000 in compute credits goes to teams pushing the boundaries of what intelligent infrastructure can do.

60 minutes of straight agentic computation. 33 concurrent agents swarming. 16m tokens blown. Less than $4.

Subagents shouldn't break the bank.

Keep your clankers clanking for less with UsePod. usepod.ai/marketplace

Our crack team of designers has been hard at work making our marketplace look nicer, but the nicest part is still the discounts!

Damn it feels good to be a gangsta...

DeepSeek V4.1 Flash is live on UsePod Released Sep 10, the smallest model in DeepSeek's new architecture family with native visual understanding. MIT license, 1M context, 552B total params with only 8B active per token. Launched and available on UsePod in less than an hour.

Only 81%? Hold my beer…

Opus 5 at 96% off. Do not fade this, anon.

Looking for private, uncensored models? Venice Uncensored 1.2 can be had now for 20% less than direct.

On the discount model side, MiniMax M3 is leading with 47% off; Deepseek v4 Flash and Pro are available for 20% off.

UsePod Discounts UsePod is the best way to buy AI inference due to the steep discounts, native crypto payments, and x402 support. Frontier models on UsePod are the cheapest around, with GPT 5.6 Sol taking the lead at almost 50% off.

Meta Muse Spark 1.3 just landed on UsePod Built for agents and long-running coding tasks. 1M token context. Uses 20% fewer tool calls and 25% fewer tokens vs. Muse Spark 1.2. Sits at #6 of 636 models on the Artificial Analysis Intelligence Index. Asks clarifying questions when stuck. Confirms before taking consequential actions. Tracks learned information across multi-step workflows instead of forgetting context mid-task. Meta's published rate is $1.25 / $4.25 standard, $0.10 / $0.20 contributor. UsePod marketplace pricing competes across providers and lands Spark 1.3 at 20% off.

Nowhere better to build.

Roll call: who has launched a @clawpumptech token for the Inference Markets track in the AnsemHack arena?

Many such cases.

Claude Fable 5.1 just dropped on UsePod Anthropic's newest frontier model--built for coding, research, and long-running agent workflows. Sets new benchmarks on agentic tasks, fixes root causes other models miss, and routes at 36% off Anthropic's published rate through UsePod.

GLM-5.3-Flash Now Available on UsePod Zhipu's newest coding model just landed on the marketplace. GLM-5.3-Flash--the open-weight multimodal model you might have tested anonymously as "Ox Alpha"--went live yesterday with MIT licensing, 1M context, and architecture built for agents that route their own inference. GLM-5.3-Flash is a 320-billion-parameter mixture-of-experts model with 18 billion active parameters, trained on 30 trillion multimodal tokens. First natively multimodal release in the GLM-5 series, processing text, images, videos, and interleaved inputs through 45 layers with hybrid attention that Z[.]ai claims delivers 3× less compute and 4.4× smaller KV cache versus the non-Flash variant.

$25,000 of $ANSEM and $10,000 worth of UsePod AI compute for the winners of the Inference Markets track in AnsemHack! Integrate at UsePod.ai and enter the hackathon at clawpump.tech/ansemhack Good luck!

Sellers on UsePod are making bank! The top ten sellers on the leaderboard have collectively sold over $3k worth of inference to the marketplace this week alone! This is AI intelligence being sold at below-market rates, getting used by AnsemHack competetors!

Homework time: If you're building on UsePod and entering the AnsemHack UsePod AI track: 1) Set your project name on the Leaderboard page 2) Follow and tag us with your logo and tell us what your idea is We will follow, retweet, and amplify competitive projects that are building on top of us. We want to help you succeed. We want the grand prize to go to a UsePod project.

Inference Capital Markets

The UsePod Leaderboard is live Rankings for who's using the most tokens and which providers are earning the most is now live. Set your name and make it public By default you're ranked anonymously. Go to your dashboard settings, set a display name, and flip the visibility switch. Your ranking shows up with whatever name you choose--your handle, your project name, your company. If you're routing serious volume or serving capacity at scale, now you can prove it. Top users Who routed the most requests this month and all-time. Measured in tokens processed, requests sent, and dollars spent through the marketplace. If you're running inference at scale through UsePod, you're on the board. Top sellers Which providers are serving the most volume and capturing the most routing traffic. Measured by requests served, tokens processed, and total volume handled over the same windows. The leaderboard that shows which upstream networks and independent GPU hosts are winning when prices get compared per-request.

OpenRouter is offering Gemini 3.7 Flash at 50% off? Hold my beer...

Wait…what???

What makes Gemini 3.7 Flash different: Google trained it with focus on "real-world software engineering and agentic benchmarks, improving issue resolution and reducing failed agent loops," plus enhanced web development capabilities that "generate higher-fidelity desktop and web application code directly from design mocks." The model ships with tunable thinking levels that let you trade quality against cost and latency depending on whether you're handling incident response or solving hard architectural problems. The benchmark gains concentrate in three places that actually matter for production use: software engineering, document-heavy knowledge work, and web development. Long-context retrieval on GDM-MRCR v2 at 128k reaches 97.0%, which means the 1M context window isn't just a spec—it's usable for the kind of whole-codebase or multi-document tasks where context actually bottlenecks the work.

Announcing Gemini 3.7 Flash availability on UsePod: Google dropped Gemini 3.7 Flash yesterday and it went live on UsePod in under an hour. A 1M token context window, 64k max output, model built specifically for coding and agentic workflows with what Google calls "our most intelligent workhorse model yet." The model delivers substantial gains on software engineering benchmarks: DeepSWE v1.1 jumped from 49.0% to 65.3%, FrontierCode 1.1 Main went from 34.4% to 43.6%, and it hit an Elo score of 1588 on Arena.ai's WebDev Arena (up from 1538 for the previous model). Google prices it at an introductory $0.75 input / $3.75 output per million tokens through December 31st, 2026, but it's available on UsePod at $0.225 / $1.13, a whopping 66% off!

xAI trained it with "longer supplemental training than Grok 4.5, with curated model-generated data for reasoning and advanced technical concepts" and reinforcement learning across kernel optimization, web development, and computer-aided design tasks. The system is built to transform "broad product ideas into working first versions" through research, architecture design, implementation, and iterative refinement--the kind of multi-turn, tool-using workflows where a 500K context window and enhanced self-testing actually matter. Four reasoning effort levels (low, medium, high, xhigh) let you trade latency for depth depending on whether you're generating boilerplate or solving hard problems.

Grok 4.6 is routable now on UsePod xAI dropped Grok 4.6 yesterday: 500K context window, 1.5 trillion parameters, built specifically for long-running agents and multi-step coding tasks. The model matches GPT-5.6 Sol Max at 61 on the AA Intelligence Index (just one point behind Fable 5 Max at 62), hits 1753 on GDPVal-AA v2, and scores 69.9% on CursorBench v3.2. xAI prices it at $2 input / $6 output per million tokens, but UsePod sells it at $1.28 / $3.84, 36% under the market rate.