Models

+1M context, vision, competitive pricing vs frontier.

−Newer — less community benchmarking than GPT/Claude.

+Built-in chain-of-thought at near-flagship price.

−Slower than non-reasoning variants.

+Optimised for tool-calling and multi-agent loops.

−Niche use-case; general tasks better on standard variants.

+Fast general-purpose at grok-3 price bracket.

−No reasoning trace; raw quality below frontier.

+Top multimodal generalist with native audio in/out.

−Highest output price in its tier.

+Quick turnaround with vision support.

−Context capped at 256k.

+Real-time X/Twitter context plus solid vision.

−Pricier per token than grok-4.x series.

+Frontier reasoning, agentic coding, 1M-token context.

−Premium price; slower throughput than Sonnet/Haiku.

+Wide 1M+ context, strong all-rounder at half the cost.

−Slower than Flash-class competitors.

+Best long-context multimodal, including video understanding.

−Preview-stage; quotas and edge-case quirks.

+Hard math, science, planning — deep reasoning.

−Expensive and slow; not for chat.

+Budget preview channel; cheapest xAI model.

−Early preview — may change without notice.

+Balanced flagship — most use cases at a sane price.

−Trails Opus on the hardest reasoning tasks.

+State-of-the-art coding for the price.

−Slow output; text-only.

+Open Chinese-English flagship with solid reasoning.

−Limited multimodal support.

+Cost-efficient chain-of-thought reasoning.

−No image input; thinks before answering.

+Open-weights with 1M context and image support.

−Slower hosted speeds than closed peers.

+Free open-weights with strong reasoning.

−Trails 5.1 on benchmarks.

+Long-context Chinese flagship with vision.

−Less common in Western tooling.

+Open agentic reasoning model.

−Text-only; smaller community.

+Native voice + image, fast multimodal pipelines.

−Outclassed on quality by the 5.x line.

+Cheap, fast coding-tuned model.

−Trails Pro on hard reasoning.

+Open MoE with near-frontier reasoning.

−Heavy to host; slower output.

+Highest throughput at this quality tier.

−Reasoning ceiling vs the Pro variant.

+Open-weights flagship — full multimodal incl. video.

−Hardware-hungry to serve at full quality.

+Mature 1M context with full audio + video understanding.

−Outclassed by the 3.x generation.

+Massive 10M token context — needle-in-haystack king.

−Quality below Maverick on dense reasoning.

+European flagship with strong vision.

−Expensive vs comparable open peers.

+MoE efficiency at near-31B quality.

−All experts must fit in memory at load time.

+Solid multimodal with real-time web context.

−Behind the frontier on raw reasoning.

+Strong open multilingual model with vision.

−Moderate speed for its size.

+Open reasoning model with visible chain-of-thought.

−Short context; slow generation.

+Tiny model with surprising reasoning chops.

−Only 16K context.

+Live web search baked into every reply.

−Text-only; quality bound by search results.

+Solid open vision-text baseline.

−No audio/video; older generation.

+Low-cost Grok variant; good for simple tasks.

−Text-only; limited reasoning depth.

+Cheap, fast Claude with image understanding.

−Limited deep reasoning vs Sonnet/Opus.

+Smallest fully multimodal model — runs on a laptop.

−Tiny context; modest quality.

+Cheap reasoning baseline for simple jobs.

−Text-only; short context.

+Tuned for retrieval / RAG and tool use.

−Behind the frontier on raw IQ.

+Cheapest OpenAI multimodal; great for high-volume tasks.

−Quality dips on complex reasoning.

+Cheap and very fast for simple multimodal tasks.

−Lower output quality on hard prompts.

+Edge-class speed with multimodal coverage.

−Smaller model loses nuance on hard tasks.

+Lightweight, easy to self-host.

−Lower quality vs the 27B sibling.

+Mature, well-optimized open-weights workhorse.

−Text-only; older generation.

+Cheap multimodal worker for high-volume jobs.

−Quality cap on hard prompts.

+Code-specialist; tuned for IDE-grade completion.

−Code-only — not a general assistant.

+Cheap, fast Grok variant.

−Lower quality; text-only.

+Fastest hosted Llama via custom LPU silicon.

−Tiny 8K context.

+Diffusion LLM with extreme inference speed.

−New architecture; less proven on hard tasks.

+Ultralight; runs on a single consumer GPU.

−Reasoning ceiling is low.

+Sub-1B model for edge devices.

−Very limited capability.

+Cinema-grade video with synchronized audio.

−Slow generation; premium tier only.

+Latest Google video model with audio understanding.

−Limited access; quotas apply.

+Strong realism with native audio output.

−Eclipsed by 3.1.

+Best creative control — motion brushes, references.

−Higher price per second.

+Fast iteration, photoreal output.

−Shorter clip length.

+Long-form, realistic motion.

−Slower generation.

+Strong stylized motion with audio.

−Less granular prompt control.

+Fluid camera moves; cheap.

−Inconsistent character identity.

+Fun edit effects, fast turnaround.

−Not photoreal.

+Open-weights video model.

−Quality below state-of-the-art.

+Open Tencent model; large-scale generation.

−Heavy compute to run.

+Real-time open-weights video.

−Lower fidelity than premium peers.

+Long, narrative-driven clips.

−Niche use cases.

+Lightweight open video model.

−Older generation; limited fidelity.

+Photoreal SOTA with sharp text rendering.

−Premium pricing; slower than 1 Pro.

+Crisp realism, tight prompt adherence.

−Eclipsed by 2 Pro.

+Best artistic style and composition.

−Discord/web-only API surface.

+Reliable text rendering inside images.

−Less photoreal than FLUX.

+Native ChatGPT integration; great instruction following.

−Tighter content rules.

+Photoreal Google quality with strong prompt fidelity.

−Restrictive content filters.

+Commercially safe; tight Adobe tooling integration.

−Conservative outputs.

+Open weights; fully customizable pipeline.

−Quality below frontier closed models.

+Full songs with vocals and structure.

−Limited fine-grained control.

+Studio-quality vocal generation.

−Smaller catalog of styles.

+Strong instrumentals plus vocals.

−English-leaning training data.

+Multilingual song output.

−Less mainstream tooling.

+Crisp production quality.

−Newer; smaller style library.

+Open audio diffusion model.

−Short max output length.

+Most expressive TTS on the market.

−Latency on long inputs.

+Low-latency real-time TTS.

−Less expressive than v3.

+Open multilingual TTS.

−Smaller voice library.

+Open expressive dialogue voices.

−One voice per checkpoint.

+Tiny open TTS — runs anywhere.

−Limited voice variety.

+High-fidelity voice cloning.

−Less English coverage than Eleven.

+Natural multi-speaker dialogue.

−Fewer base voices.