Aleph Alpha released Kolibri, an Apache 2.0 German-English model that activates 3.46B of its 78.1B parameters per token. Each token costs about as much compute as a small model, yet a team hosting it in Europe still has to fit every expert in memory.
Perspective Coverage
5 publishers
- Builder
- Builder 49%
- Operator
- Operator 36%
- Investor
- Investor 15%
Reality
- Evidence70
- Adoption
- Insufficient
- Hype gap+15
- Incentives65
- Confidence68
Google Cloud AI Research has released RRSI, an Apache 2.0 tool whose self-rewriting agent harness lifted Terminal-Bench 2.1 scores from 74.2% to 80.2%. Any team can use it commercially, though on tasks the agent never trained against the reported gain falls to 4.7 points.
Reality
- Evidence45
- Adoption
- Insufficient
- Hype gap+15
- Incentives50
- Confidence40
EDG put the C++ front end it licensed to compiler makers for 30 years on GitHub under Apache 2.0, with the C++ Alliance as its non-profit home. Toolchain teams that bought it as a closed part can now patch it in public and pay jointly for new features.
Reality
- Evidence55
- Adoption
- Insufficient
- Hype gap+10
- Incentives
- Insufficient
- Confidence60
Intel has added NVIDIA's Apache-2.0 OpenShell policy layer to its enterprise agent toolkit, putting sandboxing and kernel-level agent controls on Xeon. The feature ships switched off, so what it is worth to Intel depends on how many agent teams turn it on.
Reality
- Evidence40
- Adoption
- Insufficient
- Hype gap+10
- Incentives60
- Confidence45
Supersonic Labs' Julia 1, a 144.3M-parameter model trained for about $104, picks labels from a caller's list on an ordinary CPU. On its own 72-label Banking77 pilot it scored 64 of 100, which shows where the 20-option cap keeps it from replacing an LLM classifier.
Reality
- Evidence40
- Adoption
- Insufficient
- Hype gap+15
- Incentives45
- Confidence55
IBM's 3B, 8B and 30B dense models all get a thinking switch and native tool calling, but only the two larger ones get agentic RL, and the tuning mixture leans hard on software engineering.
Perspective Coverage
3 publishers
- Builder
- Builder 63%
- Operator
- Operator 28%
- Investor
- Investor 9%
Reality
- Evidence55
- Adoption
- Insufficient
- Hype gap+25
- Incentives60
- Confidence72
Jared Palmer's Kev now ships as 0.8B, 4B and 9B variants that score typed answer options over frozen Qwen3.5 weights, and on his own development comparison hosted Jev still scores higher than the largest of them.
Reality
- Evidence55
- Adoption
- Insufficient
- Hype gap+10
- Incentives35
- Confidence60
The Google-owned robotics unit put its real-time control, motion planning and pose estimation code under Apache 2.0 at ROSCon, along with a CNC machine tending reference design for Universal Robots and FANUC hardware.
Reality
- Evidence35
- Adoption
- Insufficient
- Hype gap+35
- Incentives55
- Confidence40
Clastix closed a 2.9 million euro seed led by CDP Venture Capital. Mistral put money in and is also a user, and its CTO says managed, multi-tenant control planes sit under the company's training and inference clusters.
Reality
- Evidence58
- Adoption38
- Hype gap+15
- Incentives70
- Confidence55
Breeze TTS 2 scores Elo 1215 and lists at $34 per 1M characters, about a third of Eleven v3's rate. Only the inference code is Apache 2.0, and commercial use of the weights needs a written agreement from BreezeBlue.
Reality
- Evidence56
- Adoption24
- Hype gap+8
- Incentives55
- Confidence58
Ternary weights at 1.76 bits put a 27B model into a 5.9GB file and let a laptop decode it at 28.1 tokens a second. The retention figure comes from Prism ML's own benchmark suite, not the table on the model card.
Reality
- Evidence38
- Adoption42
- Hype gap+34
- Incentives74
- Confidence46
Garage and RustFS are both single Rust binaries that speak S3 on one node. The documented differences are AGPL-3.0 against Apache 2.0, and four rows that Garage's own compatibility page marks as Missing.
Reality
- Evidence64
- Adoption
- Insufficient
- Hype gap+7
- Incentives38
- Confidence58
GLM-5.3-Flash leads agentic terminal work, DeepSeek V4 Flash is billed as the cheapest per token, and a 2.52B MiniCPM5-2B runs locally under Apache 2.0. The comparison flags most of those numbers as vendor-reported.
Reality
- Evidence34
- Adoption27
- Hype gap+26
- Incentives58
- Confidence41
Oathra scores an agent phone call with separate code that reads the callee's turns, and its maintainer spent this week fixing the two natural English confirmation shapes that the checker itself got wrong.
Reality
- Evidence52
- Adoption12
- Hype gap+12
- Incentives55
- Confidence55
OpenBMB's 2B-parameter model is Apache-2.0, speaks 30 languages, and serves through vLLM's OpenAI-compatible /v1/audio/speech. The real-time factor quoted for it was measured on a different backend than the one the project recommends for production.
Reality
- Evidence38
- Adoption
- Insufficient
- Hype gap+38
- Incentives42
- Confidence40
Antigravity's free tier includes Gemini 3.1 Pro and Claude Opus 4.6 with unlimited tab completions. XDA reports Opus burns credits about four times faster than Gemini, and Google moved the free tier onto credits in March 2026.
Reality
- Evidence36
- Adoption
- Insufficient
- Hype gap+12
- Incentives58
- Confidence40
Bonsai 2 27B keeps 98 percent of Qwen3.8 27B's aggregate benchmark score, up from 95 percent for PrismML's first release, and TechCrunch reports the 5.9GB file is small enough for a PC and possibly a high-end phone.
Reality
- Evidence44
- Adoption38
- Hype gap+33
- Incentives71
- Confidence55
The number came out of one harness on OpenRouter at temperature 0.9 with a 64,000-token output cap, and it holds for a bounded translation task while the same model sits mid-table on open-ended terminal work.
Reality
- Evidence57
- Adoption28
- Hype gap+19
- Incentives62
- Confidence60
Atlassian's own validator did the grading, on 25 app briefs, on the same workstation that trained the adapter. Anyone reusing the pass rate needs briefs that look like those 25.
Reality
- Evidence55
- Adoption15
- Hype gap+10
- Incentives65
- Confidence45
Amazon says the Apache 2.0 model family is generally available on Bedrock's next-generation inference engine in eusc-de-east-1, reached through an OpenAI-compatible endpoint and run by staff who reside in the EU.
Reality
- Evidence34
- Adoption
- Insufficient
- Hype gap+15
- Incentives85
- Confidence58
Earlier coverage
- Serving dockerd from WSL2 on docker.exe's own named pipe keeps Compose and Testcontainers working
Build · September 17, 2026 · 1 publisher
- Reef's commit stage decides whether a learned candidate ever replaces the version serving traffic
Build · September 15, 2026 · 2 publishers
- Hand adjudication cleared every swallowed-error flag in 120 local model generations
Build · September 15, 2026 · 1 publisher
- Edge0's 35B would need 22 GB/s of SSD bandwidth without the file cache
Build · September 11, 2026 · 2 publishers
- NVIDIA's 1.653-second H3 clip is built on a four-step adapter, with faster kernels layered on top
Build · September 7, 2026 · 1 publisher
- AWS's Agent Toolkit ships two independent halves that break in different places
Build · September 6, 2026 · 1 publisher
- AMD and NVIDIA top Hugging Face's new-model count with converted checkpoints
Build · September 4, 2026 · 1 publisher
- ChatterMate moves its agent's SQL limits out of the prompt and onto the parse tree
Build · September 4, 2026 · 1 publisher
- Nvidia's PAIR spreads one agent's model calls across whichever home PCs are idle
Product · September 3, 2026 · 1 publisher
- NeoMME runs document retrieval through one from-scratch encoder at 51 pages a second
Build · September 3, 2026 · 1 publisher
- A 15.95-million-pixel spacer div is what capped BGrid at 550,000 rows
Build · September 1, 2026 · 1 publisher
- Shipping the five-step agent loop costs between 794 and 1,729 lines
Build · September 1, 2026 · 1 publisher
- TimesFM-3 folds the promotion column into the forecast without a per-SKU model
Build · August 31, 2026 · 2 publishers
- K-Veritas signs a digest of the run so a reviewer can check the number without the data
Build · August 31, 2026 · 1 publisher
- Four-bit weights leave 6 GB on a 24 GB card for KV cache and vision tensors
Build · August 29, 2026 · 1 publisher
- ComputeLedger hashes the signer's public key into the payload to defeat swap-and-re-sign
Build · August 29, 2026 · 1 publisher
- Intel puts its Arc GPU operating knowledge inside the coding agent already installed
Build · August 28, 2026 · 1 publisher
- IBM's 20x speech gain came from deleting the decoder, not adding parameters
Build · August 25, 2026 · 2 publishers
- Three chunking revisions, RRF and a reranker, all shipped without one retrieval number
Build · August 25, 2026 · 1 publisher
- Nvidia's agent-workload lead scales with interactivity, and the second-source budget line does not
Leadership · August 25, 2026 · 1 publisher
- Laude and MIT put persistent agency in 9,900 lines of Bash, and the bill arrives with it
Build · August 25, 2026 · 1 publisher
- A 27B model reportedly beat a license check in 30 minutes. Nobody has seen the binary.
Build · August 23, 2026 · 1 publisher
- Long Horizon: Google open-sources the agent bugs that never threw an error
Build · August 22, 2026 · 1 publisher
- Instant's OpenAI exit comes with a due date: hosted apps stop on August 31st, 2027
Build · August 22, 2026 · 1 publisher
- Pydantic AI re-sends the whole run history every step, so token spend goes quadratic by design
Build · August 22, 2026 · 1 publisher
- Base Compute hands kernel tuning to agents; the carryover claim is the unmeasured part
Build · August 21, 2026 · 1 publisher
- Mojo's compiler went Apache 2.0 fifty-five days after Qualcomm's $3.92bn deal
Build · August 21, 2026 · 1 publisher
- SGLang's one-GPU Qwen3.8-27B recipe is the useful half of the release
Build · August 21, 2026 · 1 publisher
- When customers build their own features, your permission model becomes the product surface
Build · August 21, 2026 · 1 publisher
- Qualcomm buys the runtime, then hands the keys to five rival chipmakers
Leadership · August 20, 2026 · 1 publisher
- contenox stopped publishing commits and started publishing a signed tree
Build · August 20, 2026 · 1 publisher
- Two verifiers or no decision: the fix for agents that grade their own test runs
Build · August 19, 2026 · 1 publisher
- A 27B laptop model scores like a rented one, and thinks three times as hard to do it
Product · August 19, 2026 · 1 publisher
- Mojo is Apache 2.0 now. The licence was never the hard part
Build · August 19, 2026 · 3 publishers
- Test count told you nothing about the merge gate; 17 attempts to beat it told you something
Build · August 18, 2026 · 1 publisher
- Screenshot retrieval gets a one-line install, and the comparison is text embeddings
Build · August 16, 2026 · 1 publisher
- A 27B Apache-2.0 model in 17GB makes local inference a wiring decision, not a demo
Build · August 15, 2026 · 1 publisher
- A file-copy Allure adapter for Katalon, and the history IDs that make retries useful
Build · August 15, 2026 · 1 publisher
- Meta's real announcement is the split: 30B on your GPU, everything else behind the API
Build · August 14, 2026 · 6 publishers
- Qwen 3.8's Apache-licensed 27B is the one you can actually own, and its KV cache is why
Build · August 14, 2026 · 1 publisher