Skip to content
TrustList
Launches

Launches

The best new software, tools and AI models - ranked by the community. Every launch links to a full TrustList profile with real, verified reviews.

By field:Modern carsRobotics & automationAutonomous vehicles & mobility

Showing every launch, newest first. Pick a range above to rank it by community votes - weighted by each voter's trust, with editors able to reorder a launch explicitly. Featuring does not change this order: it only decides what the Featured tab shows, and that tab includes TrustList's own products.

No account yet? Email hello@trustlist.uk with launch in the subject line and we'll set it up for you - tell us the name, a one-line tagline, what it does, the date, and a link.

Sun, 22 Nov 2026

Showing 121-180 of 229 launches

  • 121
    CUA-S1-FORMSAI model

    A 2.8 MB model that decides which value belongs in which form field, in one forward pass, on the machine doing the work — 706,048 parameters, MIT licensed, with no API call and no bill.

    TL score
  • 122
    LayaAI model

    An open-weight decision model that answers typed questions with probabilities in one forward pass, about 33 ms on a T4 — and whose own model card shows where it fails while staying confident.

    TL score
  • 123

    Alibaba's streaming audio-and-video conversation model with spoken replies, tool calling and remote MCP, announced 18 September 2026; audio input costs $0.93 and audio output $1.87 per million tokens in Singapore.

    TL score
  • 124

    xAI's second-generation speech-to-text model, released 18 September 2026, for batch and streaming transcription with diarisation and up to eight channels, priced at $0.10 per audio hour batch and $0.20 streaming.

    TL score
  • 125
    Tinfield 1AI model

    Bad Theory Labs' open-weight fine-tune of Qwen3.8-Flash-Next for terminal and software-engineering agents, 177B parameters with 6.6B active and a 256K context, under the Qwen Community License; its benchmark claims are self-reported.

    TL score
  • 126
    Pareto 26.9AI model

    A blended model that runs several models on every request and keeps the best answer — 262K context, $2.50/$7.50 per million tokens.

    TL score
  • 127
    Bonsai 2 27BAI model

    A 27-billion-parameter model compressed to 5.9 GB of ternary weights, released under Apache 2.0 — small enough to run on a laptop.

    TL score
  • 128

    Alibaba's real-time interactive world model, offered on Model Studio from 17 September 2026 in Adventure, Directing and Acting modes that turn a prompt and a first-frame image into a live video world, billed per call and per second.

    TL score
  • 129
    LimiX-2AI model

    A 400M-parameter foundation model for tables rather than text: classification, regression and missing-value imputation from one pretrained checkpoint — but the licence is non-commercial.

    TL score
  • 130

    Google's low-latency speech-to-speech model for voice agents — sees video as it talks, 97 languages, $3/$12 per million audio tokens.

    TL score
  • 131

    The reasoning version of Gemini 3.8 Live — thinks and speaks at the same time; 82.6 on Artificial Analysis' speech-to-speech index, on Google's figures.

    TL score
  • 132
    JevAI model

    A decision model, not a text generator: typed answers with calibrated probabilities in 70-500 ms, $0.042 per million input tokens and free output.

    TL score
  • 133
    KoaAI model

    Salesforce's first CRM reasoning model, post-trained from NVIDIA Nemotron 3 Super and run inside Salesforce's own trust boundary.

    TL score
  • 134
    Fugu MaxAI model

    Not a model — an orchestrator that routes each task to the leanest model able to solve it, at $2/$6.

  • 135

    The capability half of Sakana's orchestration pair — 74.3 on DeepSWE, top 2 on seven of eight benchmarks.

  • 136

    Moonshot's coding and agent model for Kimi Code — close to K3 on its own account, with up to 1M tokens of context.

    TL score
  • 137

    An MIT-licensed 744B agentic model built on GLM-5.2, released as open weights by the lab behind InternLM.

    TL score
  • 138

    Singapore-based Agnes AI's agentic coding model — $0.05/$0.15 list price (currently free), with a 33B open-weights Preview.

    TL score
  • 139

    A 1M-context flash model priced by the clock — off-peak tokens cost half what peak tokens do.

    TL score
  • 140
    SWE-2AI model

    Cognition's coding model for Devin, post-trained from Kimi K3 — within a point of Fable 5.1 on FrontierCode, on its own figures.

    TL score
  • 141

    IBM and NASA's open-source foundation model for lunar remote sensing, released 10 September 2026 under Apache 2.0 and trained on about two million multimodal lunar image bundles for crater, volcanic-feature and ice-prospectivity mapping.

    TL score
  • 142

    The capability tier of OpenAI's image pair — "our most capable model for image generation and editing".

    TL score
  • 143

    The everyday tier of the same pair — "fast, high-quality everyday image generation".

    TL score
  • 144
    Mercury 2.5AI model

    A diffusion language model that writes text in parallel — 1,107 tokens a second, 260K context, $0.20/$0.75.

    TL score
  • 145
    GPT-6 AstraAI model

    OpenAI's new flagship — a premium $10/$50 tier, 1.05M context, and the first model at its own "critical" cyber threshold.

    TL score
  • 146

    DeepMind's global weather model — hourly forecasts at 5 km, live satellite ingest, up to 50% better precipitation a day out.

    TL score
  • 147

    Google's third Flash in six weeks — 90.8% on Terminal-Bench 2.1, introductory price until 31 December 2026.

    TL score
  • 148

    A security variant with looser cyber mitigations, gated to trusted defenders via the Fairwind Program.

    TL score
  • 149

    Meta's multimodal reasoning model — plus a Contributor tier at $0.10/$0.20 in exchange for training on your usage.

    TL score
  • 150

    A dated snapshot of Qwen3.8-Max, aimed at coding depth and multi-tool agent work.

    TL score
  • 151

    Anthropic's top generally available model — price unchanged, cache reads cut 75%, 52.6% on Terminal-Bench-Science.

    TL score
  • 152

    The same model as Claude Fable 5.1, with the safeguards relaxed — and no way to buy your way in.

    TL score
  • 153
    Hy4 PreviewAI model

    Tencent's open-sourced 770B mixture-of-experts model — 49B active, more than 1M tokens of context, $0.834/$2.501.

    TL score
  • 154

    The "Ox Alpha" model, revealed — trained and served on 100,000 domestic Chinese chips.

    TL score
  • 155

    The low-latency tier of the Qwen3.8 generation.

    TL score
  • 156

    Apache-2.0 reasoning with a thinking switch — 512K context, 57.0 on SWE-bench Verified.

    TL score
  • 157

    Apache-2.0, 8B, 47.7 on SWE-bench Verified — the size that fits where a 30B does not.

    TL score
  • 158
    GLM-5.3AI model

    Built to code, ready for cyber defence — 1M context, top open-source scores.

    TL score
  • 159
    Qwen3.8-27BAI model

    The small, self-hostable member of the first Max-class Qwen family to open its weights.

    TL score
  • 160

    Google's "most intelligent workhorse" — half the price of 3.6 Flash, three weeks later.

    TL score
  • 161
    Grok 4.6AI model

    Frontier intelligence for long-running agents — the cheapest model at the frontier.

    TL score
  • 162

    DeepSeek's flagship leaves preview — 1M context, 384K output, built for tool-using agents.

    TL score
  • 163

    Anthropic's most capable model — 1M context, top-tier coding & reasoning.

  • 164
    o3AI model

    OpenAI's frontier reasoning model for the hardest problems.

  • 165

    Google's reasoning-first multimodal model with a 1M-token window.

  • 166

    The Claude 5 workhorse — frontier quality at Sonnet pricing.

  • 167
    DeepSeek-R1AI model

    Open-weight reasoning model rivalling closed frontier labs — MIT licensed.

  • 168
    Qwen3.8-MaxAI model

    Alibaba’s most capable Qwen model, tuned for coding and cowork.

    TL score
  • 169
    GPT-4.1AI model

    A 1M-context GPT-4 tuned for coding and instruction following.

  • 170

    Google’s Flash-tier model family: 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber.

    TL score
  • 171
    InklingAI model

    A 975B-parameter open-weights multimodal model built for fine-tuning.

    TL score
  • 172

    Meta's open MoE flagship — multimodal, 1M context.

  • 173
    Grok 3AI model

    xAI's frontier model with strong math and reasoning.

  • 174
    GPT-4oAI model

    GPT-4 model from OpenAI.

  • 175
    FLUX1.1 ProAI model

    FLUX model from Black Forest Labs.

  • 176
    voyage-3AI model

    Voyage model from Voyage AI.

  • 177
    QwQ-32BAI model

    A 32B open reasoning model punching well above its size.

  • 178
    Qwen2.5-72BAI model

    Qwen 2.5 model from Alibaba.

  • 179

    StarCoder model from BigCode.

  • 180

    Anthropic's flagship — 1M context, 96% on SWE-bench Verified, at half the previous frontier price.

    TL score
Launches - What's new on TrustList - TrustList