Six new models, eight removals, and a sharper price split

Google’s new Gemini 3.7 Flash arrived with an input price of 0.375, an output price of 1.875, a context window of 1048576, and a score of 88. That makes it the strongest new arrival with a published score this week.

The cheaper batch version costs 0.1875 for input and 0.9375 for output, with the same 1048576 context. It has no score yet. For workloads that can tolerate batch processing, that price difference is substantial. For interactive use, the regular version is the one to compare.

Six models were added during the last 7 days, taking the tracked market to 227 models from 126 providers. The other notable arrivals were SpaceXAI’s Grok 4.6, DeepSeek V4 Pro 0813, NVIDIA’s free Nemotron 3.5 Lightning, and Upstage’s Solar Pro 4.

The new choices cover very different budgets

Grok 4.6 is the expensive, high-scoring arrival: 2 for input, 6 for output, a 500000 context, and a score of 90. Upstage’s Solar Pro 4 takes the opposite approach, priced at 0.03 for input and 0.12 for output. Its score is 71, and its context is 524288.

NVIDIA’s Nemotron 3.5 Lightning is listed as free, with both input and output priced at 0 and a 1000000 context. There’s no score yet. That makes it worth testing for low-cost or experimental workloads, but the missing score means buyers don’t yet have a quality signal in the tracker.

DeepSeek V4 Pro 0813 is listed at 1.32 for input and 3.96 for output, with a 1048576 context and no score. Its input price also doubled from 0.66 to 1.32 in the recorded price moves. Anyone evaluating it should check the current offer rather than rely on an earlier quote.

Removals matter as much as launches

Eight models disappeared from the tracked list this week. The removals include OpenAI’s GPT-5.2 Chat, Qwen3 32B, Meta’s Muse Glimmer 30B, and Z.ai’s GLM 5.2 (free). Sakana lost both Namazu entries, while LiquidAI’s LFM2.5-2.6B (free) and NVIDIA’s Nemotron 3.5 Lightning were also removed.

NVIDIA’s case is especially notable because a free Nemotron 3.5 Lightning listing arrived at the same time. That may represent a replacement, a relisting, or a provider change; the listing alone doesn’t establish which. If an application depends on a removed model, I’d confirm the model identifier and hosting provider before changing code.

Prices are moving beneath the model names

Qwen had the week’s sharpest increases among the listed moves. Qwen3 Coder 30B A3B Instruct rose from 0.07 to 0.2925 for input, a 317.9% increase. Qwen3 Coder 480B A35B moved from 0.3 to 0.975, up 225%. Qwen3 30B A3B Instruct 2507 rose from 0.04815 to 0.13, or 170%.

Z.ai’s GLM 5.2 moved from 0.49 or 0.5 to 1.4, with recorded increases of 185.7% and 180%. The duplicate entries suggest prices can change across hosts or snapshots, so a single headline price may not hold for every route.

For buyers focused on value, OpenAI’s GPT-5.6 Luna remains the standout in the tracker: a listed input price of 0.1, quality of 83.7, score of 87, and quality-per-dollar figure of 837. Gemini 3.7 Flash follows at 232.5 on that measure. Those figures are useful for screening, not a substitute for testing a model on your own prompts.

At the offer level, DeepInfra lists Mistral Nemo at 0.019 for input and 0.03 for output, the lowest listed input price. OpenAI’s GPT-5 Nano and Nex AGI’s Nex-N2-Mini are also near the bottom, while Solar Pro 4 is available at 0.03 and 0.12. My practical choice this week would be to shortlist Gemini 3.7 Flash for quality at moderate cost, Solar Pro 4 for cheap scored testing, and the free Nemotron listing only after checking its actual availability.